DOCOBEIT BESOMS
ED 195 954
AOTHOB
TITLE
INSTITOTION
SPONS AGENCy
POB DATE
CONTBACT
GRANT
NOTE
CS 005 766
Smith, Edward E.
Organization of Factual Knowledge. Teqhnical Report
No. 185.
Bolt, Beranek and Newman, Inc., Cambridge, Mass.;
Illinois Oniv., Orbana. Center for the Srudy of
Reading.
National Inst, of Education (DHEH) , Washington, D.C.;
Public Health Service (DHEW) , Arlington, Va .
Oct 80
aOO-76-0116
MH-19705
109p.
EDFS PRICE MF01/PC05 Plus Postage.
DESCRIPTORS Comprehension; *!iemory: *Models; *Prior Learning;
♦Reading Research; *Sentences
ABSTRACT
Noting that the sentence memory models formulated in
the 1970s need to be altered so as to be consistent with the fact
that people use prior knowledge to process new information, this
report discusses issues to be considered in making such an
alteration. The report focuses on three questions; (1| what
conditions lead people to elaborate and organize input facts? (2)
What are the uechaaisas behind organization? and (3) How well can
these organizational mechanisms be interfaced with the original
theories of sentence memory? The report reviews some basics of how
current sentence memory models represent and retrieve sentences and
then illustrates three conditions that lead to an elaboration of
representations. It then takes up each condition in detail, reviews
the experimental data showing that the condition doss indeed result
in an organization of the input, and delineates ths theoretical
mechanisms involved. (FL)
* Reproductions supplied by edrs are the best that can be made *
* from the original document. *
ERIC
U.S. OEPARTMENTO^ HEALTH,
eOUCATION A WELFARE
NATIONAL INSTITUTEOF
CENTER FOR THE STUDY OF READING eoucat.on
THIS DOCUMENT HAS BEEN REPRO-
DUCED EXACTLY AS RECEIVED FROM
THE PERSON OR ORGANIZATION ORrGrN.
ATINGIT POINTS OF VIEW OR OPrNIONS
SEATED DO NOT NECESSARILY 9EPRE.
SE iT OFFICIAL NATIONAL INSTrTUTE OF
EDUCATION POSITION OR POLICY
Technical Report No. 185
ORGANIZATION OF FACTUAL KNOWLEDGE
Edward E. Smith
Bolt Beranek and Newman Inc.
October 1980
University of Illinois
at Urbana-Champaign
51 Gerty Drive
Champaign^ Illinois 61820
BBN Report No. 4400
Bolt Beranek and Newman Inc.
50 Moulton Street
Cambridge, Massachusetts 02138
Preparation of this manuscript was supported by U.S. Public
Health Service Grant MH-19705 to the author, and by the National
Institute of Education under Contract No. HEW-NIE-C-400-76-0116 .
The manuscript is based on a talk given at the 197 9-8 0 Nebraska
Symposium on Motivation and will appear in the 1980 volume of
that series.
2
Organization of Factual Knowledge
1
Introduction
For the past decade^ psychologists have been intensively
studying how people represent and remember the factual
knowledge they encounter in sentence s. Perhaps the high water
mark of this research occurred in the first half of the 1970^s
with the publication of large-scale models^ such as Anderson
and Bower (1973) and Norman and Rumelhart (1975) • This work
deservedly became among the best known in cognitive
psychology. Unlike some of its predecessors in the field of
memory, the model of Anderson and Bower (1973) was sweeping in
scope yet precise in detail.
Despite their achievements, these models had a major
difficulty that rapidly became apparent to many. The problem
was that the models achieved their theoretical power by
representing only those facts that were explicit in the input
sentences, and failed to give serious attention to how people
brought to bear other knowledge that elaborated the input
facts. This was a serious omission, partly because such
elaborations are often the hallmark of true comprehension of
the input facts, and partly because these same elaborations
organize the input facts and boost their memorability. To
illustrate, suppose you read: "Herb needed a diversion" and
3
Organization of Factual Knowledge
2
"Herb looked at the movie times." In representing this
information^ you would presumably depict not only the facts
explicitly stated but also something about going to a movie.
The latter constitutes an elaboration of the input. From the
point of view of comprehension^ this elaboration is crucial
because it provides a basis for integrating the input facts
into a coherent scenario. From the point of view of memory /
it is important because the elaborated facts turn out to be
more retrievable than the uneltiborated ones.
Lately there has been a lot of attention given to the
above problem^ much of it due to the same psychologists who
produced the first generation of sentence-memory models. Thus
when Anderson (1976) proposed a revision of the original
Anderson and Bower (1973) models he devoted much concern to
how people elaborate input propositions. Similarly^ Bower
(e.g./ Black & Bower ^ 1980) ^ Norman (e.g./ Norman & Bobrow/
1976)/ and Rumelhart (e.g./ Rumelhart s Ortony/ 1977) have all
begun to study how people use whrit they already know to
organize new information. There are also many contributions
from artificial intelligence dealing with the use of prior
knowledge in representing and remembering new information
(e.g./ Adams & CollinS/ 1979; Minsky/ 1975? Rumelhart &
Ortony/ 1977; Schank & Abelson/ 1977; Winograd/ 1972).
Organization of Factual Knowledge
3
So everybody seems to agree that the early 1970s
approach to sentence memory needs to be altered so as to be
consistent with the fact that people use prior knowledge to
process new information. But in making such an alteration,
certain questions arise. The first is an empirical one:
(1) Precisely what conditions lead people to
elaborate and organize input facts?
The remaining questions are theoretical;
(2) what are the mechanisms behind organization? Is
there one basic one, or are there many?
(3) How well can these organizational mechanisms be
interfaced with the original theories of
sentence memory?
These questions form the focus of this paper.
With this as background/ a more exact agenda can be
given. The next section reviews some basics of how current
sentence-memory models represent and retrieve sentences, and
then illustrates three conditions that lead to an elaboration
of representations. In subsequent sections, we take up each
condition in detail, review the exper iment,al data showing the
condition does indeed result in an organization of the input.
ERLC
5
Organization of Factual Knowledge
4
and try to spell out the theoretical mechanisms involved. The
final section summarizes the main conclusions.
Sentence Repre sentat ions and Organizc <tiona l Possibilities
Repre se ntation a nd Retrieval in Curre nt ^fode I s of Sentence
Memory
Tlepresentation and retr ieval of single sentences .
Consider first how three different models would represent and
retrieve the proposition in the sentence "Woody Allen makes
movies." Figure 1 contains the sample representations. In
all three casvS it is assumed that the representation
corresponds to a network , where the concepts — "Woody Allen ,
"makes y " and "movies" — are depicted as nodes , and the
relations between concepts are given by labeled links between
node s .
Panel A of the figure is based on the ELINOR model
presented in Rumelhart, Lindsay^ and Norman (1972) . This
network is centered around the verb or action concept. The
node for the action, ^D;akeSr" is linked to the nodes for the
other two concepts, "Woody Allen" and "movies," while the
latter two have no direct connection. Each connection has a
label that depicts the semantic relation operative; e.g.,
"Woody Allen" is the agent of "making," "movies" is the
objec t. To see how a retrieval process might operate on this
Organization of Factual Knowledge
5
object
^- > <Movie s >
Woody
Makes Movies
ۥ ACT (1976)
Movies
Figure 1. Prepositional representations of "Woody Allen
makes movies" based on: (A) Rumelhart^ Lindsay^ and Norman
(1972) , (B) Anderson and Bower (1973) , and (C) Anderson
(1976).
ERIC 7
A. ELINOR (1972)
agent
<Wbody Allen> ^ r- ^ - ikes
B. HAM (1973)
Organization of Factual Knowledge
6
representation^ suppose that you were presented the sample
sentence as a probe and asked if you had seen it before.
According to the general ideas in ELINOR^ you would form a
reprpfaentation of the probe identical to that in Panel A, and
then use the concepts in the probe to directly access those in
the representation of the memorized sentence. Prom the
accessed memory nodes you would search for a path, i.e., a set
of labeled links^ that perfectly matches the path in the probe
representation, if you found such a path, you would say you
recogni^^ed the sentence; if you did not, you would call it a
novel se nee nee.
Panel B contains a representation from the HAM model of
Anderson and Bower (1973) • The action concept has no
privileged status here. Rather , the internal structure of the
proposition is closer to that of a phrase structure. There is
first a distinction between subject ("Woody Allen") and
predicate r and then the predicate itself is divided into a
relation ("makes") and an object ("movies"). Again we have
concepts connected by labeled links, but now some nodes are
higher-order ones that stand for complex concepts, like that
designating the entire predicate. It turns out, though, that
the terminal nodes standing for specific concepts (e.g.,
"movies") do the bulk of the work in retrieval. And the
8
Organization of Factual Knowledge
7
retrieval process for HAM looks like the one described
earlier. So if later asked if you have ever seen the probe
"Woody Allen makes movies," you would form a representation of
the probe just like that in Panel B of Figure 1, use its
terminal nodes to directly access those of the memorized
proposition, and search for a path connecting all nodes in the
memorized proposition that matches the path in the probe
representation.
Panel C of Figure 1 contains a third representation, one
based on the ACT model of Anderson (1976). Its core is the
subset of the network that relates "Woody Allen," "makes," and
a node that eventually points to the concept of "movies."
This subset is almost identical to the previously considered
HAM representation, the only difference being the label
argument has replaced the label object . The rest of the
present representation consists of a secondary proposition
asserting that what Allen makes is a subset of the class of
movies. While the latter is important for issues that
Anderson (1976) was concerned with, it turns out not to be
critical for the present discussion, and I vill ignore it in
what follows.
What cannot be ignored about ACT is its retrieval
process. The discrete search process of HAM has been replaced
Q
Organization of Factual Knowledge
8
by a continuous^ spreading activation process in ACT. When
the probe "Woody Allen makes movies" is presented^ Gach word
activates its corresponding concept in a long-term memory
network/ and activation from each of these sources spreads
a'^long the links emanating from the source. The rate of spread
on a given link increases with the strength of that lirk (an
associative-strength idea) and decreases with the number of
other links emanating from that source (an
associative-interference idea) . When the activation from the
three sources intersect^ for instance/ at the predicate node/
the path leading to this intersection can be evaluated to see
if its component links contain the same labels or relations as
those specified in the probe. If the relations are the same/
the probe would be recognized.
Representation and retrieval of sentence pairs . So much
for single sentences. From the perspective of organization,
nothing really interesting happens until we consider at least
pairs of sentences. Figure 2 contains representations of two
facts about Woody Allen: the old one about him making movies;
and a new one about him living in New York. Panel A gives the
ELINOR representation/ Panel B a simplified HAM-ACT one.
In Panel A there are two propositions connected to the
node for "Woody Allen." This should slow down the retrieval
Organization of Factual Knowledge
9
A. ELINOR (1972)
< Woody Allen>
object
Makes) — — > < Movie s>
--^..iLocation
Lives In^' > <N.y.>
B. ACT (1976)
pred.\cate
predicate
relation
Makes
Movies
Lives In
Figure 2. Prepositional representations of "VToody Allen
makes movies" and "Woody Allen lives in New York" based on:
(A) Rumelhart^ Lindsay^ and Norman (1972), and (B) Anderson
and Bower (1973) and Anderson (1976) •
Organization of Factual Knowledge
10
process in recognition. That is^ when presented with the
probe "Woody Allen makes movies^" one would again use the
concepts in the probe to access those in tht, memorized
propositions; but now the search process, which is looking for
a path matching that in the probe, will have to consider two
links off the "Woody Allen" node. If the search process is
limited in capacity, it will take longer when there are two
paths from a node than when there is only one. Therefore, the
time to correctly recognize a probe should increase with the
number of links emanating from the relevant memory nodes.''"
This same prediction follows from HAM and ACT. In Panel B
of Figure 2 there are again two links from the "Woody Allen"
node. Given the probe "Woody Allen makes movies," the three
corresponding concept nodes will be activated, where the
activation emanating from "Woody Allen" must be split between
two links. This will slow the rate of activation on the
critical link leading to "makes" and "movies." Consequently
it will takv^ longer to get an intersection that can lead to a
correction recognition.
The fan effect and interference . The three models agree
that as one learns more facts about a concept these facts fan
out from the concept node, thereby slowing any
limited-capacity retrieval process that underlies recognition.
Organization of Factual Knowledge
11
Hence recognition time for a specific fact about a concept
should slow down as we increase our knowledge about that
concept. This effect, called the fan effect, has been
extensively documented by recognition-memory studies showing
that the time to decide whether a probe sentence is Old or New
increases with the number of facts learned about each concept
in the probe (Anderson, 1974, 1975, 1976; Anderson & Bower,
197 3; Anderson & Paulson, 197 8; Hayes-Roth, 197 7; Hayes-Roth
& Hayes-Roth, 1977; King & Anderson, 1976; Lewis & Anderson,
1976; Moesher, 1979; Reder & Anderson, 1980; Shoben,
Wescourt, & Smith, 1978; Smith, Adams, & Schoor, 1978;
Thorndyke & Bower, 1974).
One reason why fanning is important is that the fan
effect on recognition latency is a common prediction of models
with different representations. Another reason why fanning is
important is that the basic idea behind it— that multiple
facts learned about a concept interfere with one another
during retr ieval— may play a role in any memory task, not just
speeded recognition. Thus an increase in fanning can lead to
an increase in recognition or recall failures if we make the
following two assumptions:
(1) People continue to search the links from a
memory node until they hit a stop rule, like
Organization of Factual Knowledge
12
"when the last n links examined have not led to
a path matching the probe ^ call the probe a New
item (if in a recognition test) or give up (if
in a recall test) " (Rundus, 197 3; Shif f r in,
1970) ? and
(2) Every time a particular link is examined it
increases in strength or accessibility and is
therefore more likely to be examined agrin
(RunduSr 1973),
«
The more facts you learn about a concept, the more likely you
are to hit the stop point before finding a target fact, and
hence the more likely you are to suffer a recall or
recognition failure* Moreover, this can happen even when you
have learned just a few facts because if you sample the wrong
link early, its accessibility will increase, and you may
continue to resample it.
Powerful as the idea of fanning is, in its unchecked form
it has a paradoxical quality (Smith et al., 1978). The idea
that the more we learn about a topic the more interference we
suffer seems at odds with our everyday experience that as we
become increasingly knowledgeable about a topic we are often
better able to answer questions about it. The way to
Organization of Factual Knowledge
13
reconcile everyday experience with the prepositional
representations and fan effects discussed is to invoke the
organizational mechanisms alluded to earlier. That is,
certain conditions lead us to alter the facts that we are
explicitly given^ resulting in a representation that allows us
to store multiple facts about a topic without substantial
increases in fanning. It is time to describe these
organizational conditions.
Conditions that Foster Organization in Sentence Memory
Facts that subdivide into distinct groups . The first
condition is chat the facts to be learned come from distinct
groups. In such a case, we may subdivide our memory
representations and thereby boost retrieval. Figure 3
illustrates this.
Panel A shows a simplified HAM-ACT representation of four
racts. (To expedite matters, relation names like subject and
predicate have bean replaced by the first letters of these
names.) The facts correspond to the two sentences previously
illustrated plus two additional ones: "Woody Allen writes
stories" and "Woody Allen dislikes LA." There are now four
links fanning off the "Woody Allen" node, and this spells
trouble for the retrieval process. Panel B shows how to get
Organization of Factual Knowledge
14
Writes Stories
B. Subdivided Network
Figure 3. Propositional representations of four facts (see
text) based on: (A) Anderson and Bower (1973) and Anderson (1976),
and (B) a subdivided network. (Both A and O are used as labels
to indicate the relation mry be thought of as either argument or
object. )
Organization of Factual Knowledge
15
rid of the trouble. The representation there has been altered
to take advantage of the natural division among the facts.
The "Woody Allen" node now leads to two subnodes: one
designating professional activities and the other^ experiences
with cities. These subnodes are in turn connected to their
relevant predicates; e.g.^ the Professional subnode connects
to the predicates concerned with making movies and writing
stories. The major elaboration is thus to create two subnodes
that were not explicit in the input factS/ and to insert each
subnode between the relevant subject and predicate nodes in
what is called a subdivided network .
Facts that can be integrated by prior knowledge . To
illustrate this condition^ consider the HAM-ACT
representations in Figure 4. Panel A represents two
sentences: "Herb needed a diversion" and "Herb waited in
line"? Panel B represents the same two sentences plus two
additional ones: "Herb went to a movie" and "Herb V ought
popcorn." There is an increase in fanning off the "Herb- node
from two to four as we move from Panel A to Panel B. This
means that the models we reviewed earlier would predict it
takes longer to correctly recognize "Herb waited in line" if
one learned the four sentences in Panel B rather than only the
Lwo in Panel A. Intuition suggests otherwise. The four facts
Organization of Factual Knowledge
16
A. 2 Pacts
Figure 4. Prepositional representations based on
Anderson and Bower (1973) and Anderson (1976) for: (A)
two unrelated facts and (B) four integrated facts (see
text) •
Is
Organization of Factual Knowledge
17
in Panel B seem to make up a coherent unit while the two in
Panel A are more difficult to integrate. Here our
sentence-memory models seem to be missing a critical
point: facts integratable by some prior knowledge (like
knowledge about going to a movie) may not function as
independent propositions in memory.
Facts containing perfectly correlated predicates . The
condition of interest is illustrated by the sentences in Table
1, On both the left- and right-hand sideS/ there is one fact
about John^ one about Ed^ and three each about Woody and Mel,
Furthermore / for both the sentences on the left and those on
the right- each predicate is used twice^ and the trio of
predicates attributed to Woody or Mel are not readily
integrated by any salient packet of pr:or knowledge. What
then is the difference between the two sentence sets? The
predicates on the left are perfectly correlated whereas those
on the right are not. Given the facts on the left/ if someone
makes movie s^ that same someone was born in Brooklyn and went
to California; or if someone visited Virginia, that is all he
did. In contrast/ for the facts on the right, if someone
makes movies, he may have been born in Brooklyn (like Woody)
or he may net have (like Mel) . Perfectly correlated
predicates seem to provide some structure to the input facts.
Organization of Factual Knowledge
18
Table 1
Sentences with Predicates that
Vary in Correlation
Perfectly Correlated
John visited Virginia
Ed visited Virginia
Less than Perfectly Correlated
John visited Virginia
Ed was born in Brooklyn
Woody was born in Brooklyn
Woody makes movies
Woody went to California
Mel was born in Brooklyn
Mel makes movies
Mel went to California
Woody was born in Brooklyn
Woody makes movies
Woody went to California
Mel visited Virginia
Mel makes movies
Mel went to California
^0
Organization of Factual Knowledge
19
And people seem to use this structure: though the variations
in fanning are identical in both sets of sentences, only the
sentences on the right produce a fan effect on recognition
latencies (Whitlow, Medin, & Smith, Note 1).^
Subdividing Faets from Distinct Groups
The first task is to present empirical evic«once about how
facts memorized from distinct groups facilitates retrieval.
Wte start with research on the fan effect and then move on to
other empirical phenomena. After that, we will discuss
theoretical mechanisms.
Empirical Evidence
Fan effects on recognition latency . A few experiments
have dealt with recognition latencies for memorized facts from
distinct groups. One of the simplest is by McClosky (Note 2).
He first had subjects learn from one to six facts about
various people identified by occupation terms, such as the
tailor . For each occupation term, some facts concerned
animals, the rest countries. And for each occupation term,
McClosky manipulated the fan level of the animal facts
independent of the fan level of the country facts. This is
illustrated in Table 2. For the tailor there are five facts
about animals but only one about countries, while for the
chemist there is one fact about animals but five about
21
Organization of Factual Knowledge
20
countries. After memorizing these facts, subjects were given
an Old-New recognition test. The memorized facts were
intermixed with a like number of distractors, where each
distractor was constructed by repairing the occupation term
from one learned sentence with the predicate of another, e.g.,
"The tailor likes Canada" (see Table 2). The subject's task
was to decide as quickly as possible whether each
sentence — referred to as a probe — was Old (on the memorized
list) or New (a distractor) . The data of major interest were
the average times needed for correct Old and New decisions.
To appreciate the results of the above experiment, note
that any probe contains both an occupation and an object term
from the memorized list. The occupation term is characterized
by two fan levels, one designating the number of learned
animal facts, the other the number of learned country facts.
The object term of the probe, however, tells the subject which
of these two sets of facts is relevant; e.g., if the object
names an animal, only the animal facts are relevant. Thus,
though a probe has two fan levels, one may be designated a
relevant fan, the other an irrelevant fan. To illustrate with
the items in Table 2, if the probe was "The tailor likes
wolves, "the relevant fan would be five and the irrelevant fan,
one. If McClosky's subjects used the object term of the probe
'^2
Organization of Factual Knowledge
21
Table 2
Example of Sentences Used in McClosky (Note 2)
Facts
about
Animals
The tailor likes wolves
The tailor likes rabbits
The tailor likes bears
The tailor likes tigers
The tailor likes pigs
The chemist likes wolves
Pacts
about
Countries
The tailor likes Portugal
The chemist likes Portugal
The chemist likes Canada
The chemist likes England
The chemist likes Brazil
The chemist likes Italy
Organization of Factual Knowledge
22
to decide which set of memorized facts was relevant^ their
recognition latencies should have systematically increased
with the relevant fan but not with the irrelevant fan. This
is roughly what McClosky found. Recognition latency increased
about 370 msec as the relevant fan increased from one to five,
but increased only about 100 msec as the irrelevant fan varied
over this same range. While the 100 msec increase may suggest
subjects were considering the irrelevant facts, two points
mitigate against this: the 100 msec increase did not reach a
conventional level of statistical significance; and part or
all of the increase may reflect the time needed to decide
whether the object term names an animal or country. All
things considered/ McClosky's results suggest that people can
sometimes subdivide their knowledge and use information in the
probe to direct their search to the relevant subgroup.
In the above study, the subdivision was based on a
semantic aspect: the object term named either an animal or a
country. Anderson and Paulson (1978) looked at a different
baisis for subdivision. Using a paradigm like that described
above, they showed that if some facts about an object are
presented pictorially while others are presented as verbal
descriptions, subjects may use this difference in mode of
presentation as a basis for subdivision. When Anderson and
S4
Organization of Factual Knowledge
23
Paulson's subjects were given a recognition test, if the probe
was a verbal description of an object, for example,
recognition latency increased substantially with the number of
verbally presented facts about that object but only minimally
with the number of pictorially presented facts about the
object.
The above studies have two limitations. First, the only
bases for subdivision that have thus far been demonstrated are
simple ones — the semantic category of a probe word and the
modality of presentation. If subdivision is confined to such
simple aspects, it could not play much role in real-life
memory situations and hence could not be the only means of
organization used. Second, though the above results suggest
people can restrict their search to the subgroup deemed
relevant by the probe, we will soon have cause to question
whether search processes in irecognition are usually this
selective .
Free recall of categorized lists . The previous studies
make excellent contact with models of sentence memory because
they focus on the fan effect. But while these studies are
analytic, they provide too narrow a view of subdivision.
Subdivision can have striking effects on recall, both the
amount recalled and the structure c the recall. Thest
Organization of Factual Knowledge
24
effects have recently been demonstrated in prose recall (Black
& Bower r 1980) ^ but they have been most extensively documented
in studies dealing with the free recall of word lists.
Those concerned with free recall of word lists long ago
discovered that recall improves when the words are drawn from
a few semantic categories. Suppose subjects are presented a
list of 40 words. They will recall more if the list consists
of five instances froin each of eight semantic categories — just
the instance Sr not the category names — than i:c all words are
semantically unr«»lated (e.g.^ Cohenr 1963? Puff, 1970). This
effect depends pr-vrtly on subjects being aware of the
categorical sti cv.ire of the list 5;t the time of input (Gofer,
Bruce, & Reicher, 1966), which suggests the effect depends on
setting up a certain kind of representation. The obvious
possibility is a representation that is subdivided according
to categories (Bousfield & Cohen, 1953). Figure 5 presents
an example. Though the category terms — animals , c ountries ^
etc. — did not appear in the input list, they have been
inserted in the representation as subnodes and are connected
to nodes for words that did appear in the input.
This representation seems consistent with three important
findings about recall from categorized lists.
o
ERIC
Organization
of Factual Knowledge
25
Figure 5. Example of a subdivided representation for a
categorized list.
k7
ERIC
Organization of Factual Knowledge
26
(1) Recall is clustered by categor ies . Subjects
recall a number of instances from one category
in success ion^ then a number of instances from
another category^ and so on (Bousf ield, 1953) •
This suggests that subjects retrieve a category
subnoder search its links to instances^ and move
on to the next subnode.
(2) Recall is "some or none . " Typically either
several instances of a category are recalled or
none are (Cohenr 1965) , This suggests that if
one cannot get to a particular subnode^ there is
no other path to its instances.
(3) Category cueing aids recall . If recall is
substantially less than perfect, giving subjects
the category terms as cues will enable them to
retrieve some of the missing items (Tulving &
Pearlstoner 1966). The. cues apparently allow
subjects to retrieve subnodes they missed in
their initial recall.
Theoretical Meclianisms
Retrieval processes for free recall . To illustrate the
mechanisms involved, consider Figure 6. It contains some
^8
Organization of Factual Knowledge
27
facts that a devotee of Woody Allen might have stored. The
top node designates the concept "Woody Allen." Then there are
two levels of subnodes. The first distinguishes "Personal"
and "Professional Life," while the nature of the subnodes at
the next level depend on whether they are dominated by
"Professional" or "Personal-Life" subnodes. Subnodes under
"Personal Life" designate periods or eras of life (Kolodner,
1978)— like "Early Childhood," "Boyhood," "Marriage,"
etc. — whereas subnodes under "Professional Life" designate
different occupational and artistic roles — like "Gagwriter,"
"Storywr iter, " and "Filmmaker."
Consider how the information in the network might be
retrieved during free recall. If the possessor of the above
network were asked to say everything he knew about Woody
Allen, he would presumably enter at the top node and traverse
either the link to the "Professional" or "Personal-Life," with
the strength or accessibility oi^ these links determining which
one is chosen (Rundus, 1973; Shiffrin, 1970). Assume he took
the path to "Professional Life." Then he would traverse one
of the links leading to a more specific subnode, e.g.,
"Filmmaker" (the choice again determined by link strength),
and start searching paths from this specific subnode, emitting
each fact he found, e.g., "Woody Allen's films include Annie
Woody AUen
Early
Childhood
Personal Life
Professional Life
Born 1936
Boyhood Marriage Other
Relations
Born in
Brooklyn
Raised Little
in League
BrooUyn
Named
Mien KonigBberq
Marshall
Louise \ Bricknan
Lasser
Twice
Divorced
Gagwriter Storywriter Filmmaker
\
Sid Cesar
Started in
High School
New
Yorker
Without
Feathers
Getting Even
Analysis
Bananas^
Interiors Love
Pla;[ it Again Death
San
Annie Hall
Manhatten
to
ob
Figure 6, Example of a subdivided network for knowledge
about Woody Allen.
0
N
P>
ft
H-
0
9
0
Hi
P
0
ft
pi
«
I
H
30
ERIC
(D
? 31
Organization of Factual Knowledge
29
Hall and Manhattan ," Having done what can ^ith the facts
under a specific subnode^ our respondent would presumably find
his way back to the higher fiubnode that was activated,
"professional Life," trace a path down to another specific
subnode, e.g., "Gagwriter," and start searching pathr from
there to specific facts. Once he has done his utmost with the
accessible subnodes under "Professional Life," our respondent
would move on to "Personal Life" and the specific subnodes it
dominates.
This scheme is consistent with the free-recall findings.
Following the process just outlined, our respondent would
cluster his recall by specific subnodes (e.g., he would recall
Allen^s movies in one group), as well as by higher-level
subnodes (e.g., he would recall Allen^s movies closer to
Allen^s published stories than to facts about Allen^s
boyhood) . His recall should also have a some-or-none
character, e.g., he would either not mention anything about
Allen^s marriages or mention most facts he has stored about
them. And should our respondent fail to emit anything at all
about a particular subnode like "Gagwriter," reminding him
that Allen was once a gagwriter might bring forth the relevant
facts.
Organization of Factual Knowledge
30
Co nsiderations of efficiency . The above retrieval scheme
is extremely efficient because each node has a relatively
small number of links fanning off of it. We were able to
depict over 20 facts about Woody Allen while keeping the
maximum fan off a node down to six. The largest nur/iber of
links one would ever have to traverse is eight (for a question
about films made)^ while the smallest is four (for a question
about gagwriting). High efficiency could be maintained even
with a substantial increase in the number of facts by
increasing the number of subnodes at each level and/or the
number of levels.
However r research on the recall of categorized lists
suggests a limit to the amount of subdivision possible.
Handler ^s (e.g.r 1967) studies indicate that recall is maximal
with 5i2 categories or nodes per level; if more than this are
used to divide up a fixed number of facts, recall starts to
decline. Since there is no reason to think otherwise, we
assume this 5±2 limit might hold at all levels. What about a
limit to the number of levels? As no one seems to have worked
with categorized lists using more than three levels, it is
possible that this factor is also governed by some small
number. If it is again about five, an optimal subdivided
network, five levels with five nodes per level, could
33
Organization of Factual Knowledge
31
represent 3^225 facts without requiring the retrieval process
to ever inspect more than 25 nodes. This remarkable
efficiency may account for why hierarchical representations
have been shown to be such powerful recall aids in general
(Bower, Clark, Winzenz, & Lesgold, 1969? Nelson & Smith,
1972), and why they are so widely used as storage devices in
computer systems (where they are referred to as discrimination
nets) .
Retrieval processes for recognition . For recognition,
there is in principle no need to search the entire network*
To illustrate, given the probe "Was Woody Allen ever married
to Louise Lasser?", our respondent could enter at the top of
the network, use the probe to get to the "Personal-Life"
subnode, and then again use the probe to get to "Marriage"
(Note that to use the probe to get to "Personal-Life," our
respondent must know that marriage pertains to personal life) .
Under this view, the search process is selective, in the sense
that it uses information from higher-level subnodes to select
the appropriate lower -level ones.
Problems for this view arise, however, if we alter our
probe slightly to, "Was Woody Allen ever involved with Louise
Lasser?" Now the analysis of the probe that gets our
respondent to "Personal-Life" must be quite complex. It
Organization of Factual Knowledge
32
cannot simply use "involved with" as an access condition for
"Personal-Life^ " for the question "Was Woody Allen ever
involved with Sid Caesar?" will get our respondent to the
"Professional-Life" subnode. It seems that to get to the
"personal-Life" node for the Louise Lasser question but not
for the Sid Caesar oner we have to consider the fact that
Woody Allen is a notoricus heterosexual r thereby making it
plausible that "involved w.lth" can be given a romantic reading
with Louise Lasser but not with Sid Caesar. But Allen^s
heterosexuality is the kind of fact that is presumably
represented at some lower-level subnode r so how can we have
access to it while still working at a top-level subnode? More
generally^ selection of a higher-level subnode may sometimes
rest on information at lower-level nodes, which is at odds
with the idea of a selective search where one only accesses
lower-level nodes by first going through higher-level ones*
It seems, then, that search processes in recognition are
considering lower-level nodes at the same time as higher-level
ones* To illustrate with the above example, we seem to search
up from the nodes for Louise Lasser or Sid Caesar at the same
time we search down the nodes of the hierarchy.
Considerations like these in a different domain led
Anderson (1976, Chapter 8; King & Anderson, 1976) to reject
Organization of Factue;! Knowledge
33
the notion of selective search altogether ^ i^nd to opt for a
spreading activation process that starts at the probe concepts
and then searches blindly through a network. This seems to be
a reasonable move^ but there are two problems with it that
have to be faced. Firsts we have to reconcile the lack of
selective search in recognition with the idea that search
appears directed in free recall/ i.e., enter at the top of the
hierarchy and search systematically through it. (Direction is
a necessary component for selection.) This can be done by
noting that there is typically only a single retrieval cue in
free-recall — in our free-recall example/ the only cue was the
name "Woody Allen" — and this cue permits access to only the
top of the network. In contrast/ a recognition probe
typically contains multiple retrieval cues — e.g./ "Woody
Allen/" "involved with/" and "Louise Lasser" — thereby
permitting simultaneous access to multiple parts of the
network. Under this vieW/ which is essentially due to Tulving
(1974)/ people use whatever retrieval cues they can and
so-called directed-search is what happens when they are forced
to work with a single cue.
The second problem is that the fan experiments reviewed
earlier/ McClosky (Note 2) and Anderson and Paulson (1978)/ do
provide evidence for selective search in recognition^ which of
3G
Organization of Factual Knowledge
34
course contradicts the generalization that such a search is
not used in recognition. A resolution here may hinge on
something mentioned earlier: the fan experiments in question
used very simple bases for subdivision, such 6 the r lanr
category of a probe word. Perhaps selective s^-^^ch is used
when the basis for selection is easily computed from the
prober like determining whether the last word of the probe
names an animal or country, but is not used when such
computations become at all complex, like determining "involved
with Ixjuise Lasser" means something about romance.
The status of subdivided networks as a r organization
device . Let us summarize the main points made above. Free
recall data provide good evidence that we car subdivide our
knowledge, and that the concomitant reduction in
fan-level-per-node facilitates the retrieval process. Since
we can subdivide when preparing for free recall, it seems
likely that we can also do so for recognition. In
recognition, however, subdivision also has the potential to
permit a selective search (as well as a reduction in
fan-level-per-node) , but such selectivity may only occur in
certain simple cases.
So at this point, subdivision-without-selectivity seems a
reasonable organizational devica, primarily because of its
57
Oirganization of Factual Knowledge
35
reduction in fan level. There is, however, a cost to
subdivision that places limits on how widespread a device it
can be. Dividing our knowledge into different chunks ignores
existent relations between facts stored under different
subnodes. Since people know these relations and use them in
answering questions, subdivided networks cannot be the only
way we represent substantial bodies of knowledge.
can again illustrate with the Woody Allen network.
Under the "Personal-Life" node, we had a subnode for romantic
relations that was connected to facts about Allen's relation
with Diane Keaton, while under "Professional-Life" we had a
"Filmmaker" subnode connected to facts about Allen's movie
Annie Hall . But as any devotee of Allen knows, his relation
with Keaton formed the basis for Annie Hall . So to be true to
our knowledge base, we need some sort of connection between
these disparate subnodes. One way to do this is to insert a
link between the film Annie Hall and the relevant facts about
Allen's relation to Keaton. But this move can substantially
increases the number of facts fanning off the "Annie Hall"
node, yet the whole point of subdivision is to keep the
fanning down; An alternative is to add a fact to the
"Annie Hall" node, namely that it was based on Allen's
relation to Keaton. This will increase the fanning off "Annie
Organization of Factual Knowledge
36
Hall" by only one fact. But this move is not really faithful
to the knowledge of a Woody Allen fanatic who presumably knows
how various aspects of Allen's personal relations mapped onto
different aspects of the film in question. That is, part of
what is known here is how one structure- maps onto another.
More generally, part of a rich knowledge base about any topic
consists of relations between seemingly disparate facts, and
subdivided networks seem more disposed to keeping such facts
apart than to depicting their • subtle connections.
Fact3 Integratable by Prior KnowJ.edge
Empirical Evidence
Again we first consider findings about fan effects, then
take up results with other memory measures, and lastly
consider theoretical mechanisms.
Fan effects on recognition latency . A few recent
experiments demonstrate that learning new facts about a topic
causes little fan effect when the propositions are
integratable. in the first set of studies (Smith, Adams, &
Schoor, 1978), subjects learned either two or three facts
about a person designated by an occupation term, such as the
banker . Some subjects learned facts that were -easily
integratable by prior knowledge, like those in the top of
Table 3. The two facts about the bank -^r fit with what we know
Organization of Factual Knowledge
37
about christening a ship, the three facts about the accountant
are consistent with knowledge about playing a bagpipe. The
remaining subjects learned facts that were not so
integratable^ as illustrated by the items in the bottom of
Table 3.
For both the integrated and unrelated facts in Table 3,
the fan off the "banker" node is two and that off the
"accountant" is three. This means that current models of
sentence memory would expect comparable fan effects on
recognition latency for both kinds of facts. When subjects in
the Smith et al. study were given a recognition task after
learning the facts, however r there was a substantial fan
effect wlUli the unrelated items but not with the integrated
ones.
Apparently subjects given the facts in the top half of
Table 3 used their world knowledge about ship christenings and
playing a bagpipe to integrate the facts. That world
knowledge was indeed activated showed up in other findings by
Smith et al., specif ically, findings concerned with the
distractors in the recognition task. Most distractors were
formed by repairing the occupation term from one learned
sentence with the predicate of another (call these repaired
distractors) ; some distractors, however, were formed by
Table 3
Example of Sentences Used in Smith, Adams, and Schoor (1978)
Integrated Facts
The banker was chosen to christen the ship The accountant played a damaged bagpipe
The banker broke the bottle The accountant produced sour notes
The accountant realized the seam was split
Unrelated Pacts
0
n
pi
The banker was asked to address the crowd The accountant painted an old barn 3
I*
N
The banker broke the bottle The accountant produced sour notes J.
The accountant realized the seam was split §
0
Hi
U) HI
00 0
(+
c
p»
H
ERIC
Organization of Factual Knowledge
39
changing one word in a learned sentence so that it remained
consistent with the relevant world knowledge (call these
related distr actor s) . To illustrate with the distractors for
"the banker" (see Table 3), a repaired distractor would be
"The banker realized the seam was split," while a related one
would be "The banker broke the champagne bottle." The
findings of interest were that subjects who learned integrated
facts responded slower and made more errors on related than
repaired distractors (presumably because related distractors
were consistent with the accessed world knowledge), while
subjects who learned unrelated facts did just the reverse.
Hence one indication of the use of world knowledge is the
difficulty of rejecting distractors consistent with the
knowledge.
The relative lack of a fan effect with integrated facts
has been replicated by Moesher (1979) and Reder and Anderson
(1980). Both studies, however, revealed constraints on the
power of integration to offset the fan effect. Moesher (1979)
demonstrated that integrated facts are insensitive to fanning
variations only when the facts are presented close together
during learning. For example, if successive facts about
christening a ship are separated by five irrelevant sentences,
the christening facts will behave like unrelated ones. This
4P
Organization of Factual Knowledge
40
suggests that at least a couple of the integratable facts must
be in active memory at the same time in order for the relevant
world knowledge to be accessed.
Reder and Anderson ^1980) showed that the fan effect is
diminished with integrated facts only whem the distractors are
not always consistent with the world knowledge needed for
integration. To illustirate^ suppose subjects first learned
facts about a particular person that all dealt with skiing and
then were given a recognition test. If the distractors on the
recognition test (a) used recombined subject and predicate
terms and (b) were always consistent with skiing^ there was as
substantial a fan effect as occurs with unrelated facts. A
plausible interpretation of this finding goes as follows. The
representation that results when world knowledge is used to
integrate input facts cannot adequately discriminate between
the input and novel facts equally consistent with the
knowledge^ so when all distractors are consistent with the
world knowledge^ subjects are forced to use an unelaborated
representation of the input.
In addition to varying the number of facts integratable
by some packet of world knowledge or theme (e.g., skiing,
washing clothes), Reder and Anderson also varied the number of
themes learned about a particular person. Thus, subjects
^4
Organization of Factual Knowledge
41
might have learned three skiing facts and one clothes-washing
fact about a fictitious character named Arnold (i.e,, two
themes about Arnold)^ but only three skiing facts about a
character named Bruce (one theme about Bruce) • Even in
conditions where recognition latency was unaffected by the
number of facts within a theme ; latency did increase with the
number of themes learned about a person.
Experiments on recall and recognition accuracy . Numerous
studies show that when subjects use their prior knowledge to
integrate some presented facts.^ recall accuracy of the learned
facts is increased but at the price of intrusions that are
consistent with the relevant knowledge. In a similar vein,
recognition studies show that integration via previous
knowledge leads to batter recognition accuracy of the learned
items but at the cost of more false alarms to distractors
consistent with the knowledge. Representative studies of each
type are described below.
!• Integration and recall . Bransford and his colleagues
(e.g., Bransford & Johnson, 1973; Bransford & McCarrell, 1974)
have performed several experiments that take the following
form:
EKLC
Organization of Factual Knowledge
42
(1) All subjects are presented some facts that
appear to be unrelated;
(2) One group of subjects are also presented a clue
specifying a packet of world knowledge that can
be used to integrate the presented facts, while
the remaining subjects receive no such clue; and
(3) Subjects given the clue rate the presented facts
as more comprehensible, and recall more of them
on a subsequent recall test*
We can illustrate vfith the Bransford and Johnson (1973)
study* The seemingly unrelated facts comprised an obscure
paragraph, whose first few lines were:
The procedure is quite simple* First you arrange
things into different groups. Of course, one pile
may be sufficient, depending on how much there is to
do* If you have to go somewhere else due to lack of
facilities, that in the nwct step; otherwise you are
pretty well set*
Subjects given the clue "washing clothes" at the time of input
rated the paragraph more comprehensible and subsequently
recalled more propositions from it than subjects lacking the
46
Organization of Factual Knowledge
43
clue. The power of the clue resides in its ability to access
knowledge about the actions typically involved in washing
clothesr where this knowledge can then be used to elaborate
and integrate the input propositions.
The above example shows the beneficial effects of
integration but says nothing about its costs. The latter has
been demonstrated by Bower r Black, and Turner (1979). They
had subjects read stories about recurrent, stereotyped
situations, like going to a restaurant. Subjects presumably
utilized their world knowledge about such situations in
understanding the input stories, and in a subsequent recall
test, the bulk of the intrusions were consistent with the
world knowledge presumably accessed.
2. Integration and recognition . Some experiments on
recognition have used a cueing vdria.ion similar to that
employed in Bransford's studies. in Dooling and Lachman
(1971) , for example, all subjects were presented the following
obscure paragraph:
With hocked gems financing him our hero bravely
defied all scornful laughter that tried to prevent
his scheme. "Your eyes deceive," he had said, "an
egg not a table correctly typifies this unexplored
47
Organization of Factual Knowledge
44
planet." Now three sturdy sisters sought proof r
forging along r sometimes through calm vastuess. yet
more often over turbulent peaks and valleys. Days
became weeks as many doubters spread fearful rumors
about the edge. At last from nowhere, welcome
winged creatures appeared signifying momentous
success »
One group of subjects was given the clue at the time of
input that the paragraph was about "Christopher OolumbuSr"
while the remaining subjects made do with no clues> At some
later point, all subjects were given a recognition test. It
included old sentences from the above paragraph intermixed
with distractorsr where some distractors were related to the
Columbus saga. Subjects given the clue correctly recognized
more old sentences than their non-clu€id counterparts, but the
clued subjects H^ire also more likely to false alarm to the
related distractors.
Theoretical Mechanisms
Two different kinds of mechanisms need to be considered.
The first involves an extension of subdivided networks. The
second focuses on some new processes, namely inferences made
during comprehension.
Organization of Factual Knowledge
45
Subdivided networks . The ideas here were developed by
Reder and Anderson (1980) to account for why the fan effect on
recognition latency is reduced if all facts learned about a
character can be integrated by some prior knowledge. To
extend an earlier example, if subjects already know that "The
banker christened the ship" and "The banker broke the bottle,"
then learning that "The banker did not delay the trip" does
not slow them down in answering questions about "the banker."
According to Reder and Anderson, when learning the above
facts, subjects presumably set up a subdivided network like
that in Figure 7. The "banker" is the top node,
"ship-christening" the only subnode, and the three specific
predicates comprise the bottom nodes. This looks like the
subdivided networks we considered previously. But there is
something new here. In addition to the "ship-christening"
subnode being attached to the three specific predicates, it is
also associated with concepts relevant to ship christening,
such as "bottles," "trips," and "champagne." These connections
constitute the subjects' prior knowledge about ship
christening, and they play a critical role in Reder and
Anderson's subnode -activation hypothesis. Specifically, when
a probe is presented, e.g., "The banker broke the bottle,"
there is activation at the "banker" node as well as at the
concept nodes representing the relevant prior knowledge.
Organization of Factual Knowledge
46
Activation from the latter nodes travels directly to the
subnode along the pre-existent paths^ while activation from
the "banker" node goes to the subnode along the link created
in the experiment. So the subnode is the first likely point
of an intersection of activation^ and such an intersection is
assumed to be sufficient for recognition^ i.e., sufficient for
subjects to respond "old" to a probe.
ThuS/ even though search is not selective (all probe
concepts are activated simultaneously) ^ and even though there
is a substantial fan off of the subnode ^ the
subnode -activation hypothesis is consistent with
organizational effects. There is no fan effect because the
search process need not examine the learned predicates.
Related distractors {e.g., "The banker broke the champagne
bottle") are difficult to reject because they contain terms
that activate ti\e prior knowledge concepts and consequently
can lead to a spurious intersection at the subnode. If all
distractors are related, aa in some conditions of Reder and
Anderson (1980), subnode activation is no longer a useful
indicator of what facts were actually presented; hence,
subjects will be forced to search the specific facts, and the
fan effect should reappear. Lastly, the hypothesis explains
why recognition latencies increase with the number of themes
sg
Organization of Factual Knowledge
47
/
Banker
/
/ y
Bottles
/ y
Trip
Ship-Christening
A/^^ Champagne
istenina.^ '
Chosen to
Christen ship
Broke the
bottle
Didn't delay
the trip
Figure 7. A subdivided network for three integrated
facts. Dashed lines indicate prior associations to concepts
(see text) •
Organization of Factual Knowledge
48
learned about a character. Each theme requires a different
subnode (as well as different set of prior -knowledge
concepts) 9 and the more subnodes, the less activation to any
one of them from the top node of the network. So the number
of relevant themes slows recognition because it slows the rate
oi! top-down activation^ while the number of facts-per-theme
has no effect on recognition because such facts need not be
examined if the distractors are unrelated to any theme.
The virtue of tha above hypothesis is that it explains
the recognition results for integratable facts by the same
kind of mechanism used to account for results with facts from
distinct groups. Thus only one basic mechanism is needed to
account for two seemingly disparate kinds of organizational
effects^ where this mechanism is readily inter faceable with
Anderson'^s (1976) existent theory of sentence memory,
Ther e ar e ^ however 9 two 1 imitations to thi s subnode
approach. The first is that while it can account for
variations in recognition accuracy as well as in recognition
latency^ it is unclear how it would explain the comparable
recall results. In the preceding accounts of fan effects, we
assumed connections between a subnode and the specific
concepts previously associated with it; but to account for
recall, we need connections between a subnode and entire
EKLC
52
Organization of Factual Knowledge
49
propositions^ since in a recall test people emit full
sentences^ not single words designating specific concepts. We
will not dwell on this because it is unclear whether or not
the needed modification of the subnode approach can be easily
made. The second difficulty with the approach is that it
focuses exclusively on memory for integrated facts and ignores
processes involved in the comprehension of such fjacts. This
is problematic because it may turn out that the nemory
phenomena obtained with integrated facts are being mediated by
comprehension effects. This leads us to a second kind of
theoretical mechanism.
Interconnecting inferences by schemas .
1- World knowledge and comprehension . We can begin by
expanding on the above suggestion that the effects of prior
knowledge on memory may be mediated by the effects of such
knowledge on comprehension. Figure 8 provides an abstract
illustration of this. Suppose a reader is presented two input
facts and accesses some relevant world knowledge to aid in
understanding them. The world knowledge , will be used to
generate inferences. Some inferences will establish direct
relations between the input facts ^ illustrated in Figure 8 by
the link between Input Facts 1 and 2. Other inferences will
result in inferred facts that yield a multi-link relation
Organization of Factual Knowledge
50
between the input facts; this is illustrated by Inferred Pact
2, which creates a two-link relation between the input facts.
Once a fact is inferred, it may lead to other inferences that
create still other multi-link relations between the input}?
thus Inferred Fact 1 leads to Inferred Fact l"*, which in turn
leads to Inferred Fact 2, thereby creating a four-link
relation between the input facts. Lastly, there will Le
inferred facts that do not result in any connection between
the input, as illust:rated by Inferred Fact 3.
The result of all this inferencing is a representation
that goes far beyond the input, one that shows enablement and
causal relations between propositions and that can be used to
answer all sorts of questions about the input. The
construction of such a representation is what many people mean
by comprehension ^ Under this account, the major purpose of
coutacting world knowledge during reading is to facilitate
comprehension. But, and this is the critical point, note that
in constructing this representation many of the inferences
have interconnected the input facts, and interconnections per
se are good for memory retrieval > For having found one fact,
the retrieval process can follow the path to a second one. So
a side benefit of the inference process is that it facilitates
subsequent retrieval. Hence, the claim that many effects of
Organization of Factual Knowledge
51
Figure 8. Abstract illustration of use of v;orld
knowledge in understanding two input facts.
Organization of Factual Knowledge
52
world knowledge on memory are mediated by effects on
comprehension.
Another consequence of accessing world knowledge is
represented by the dashed line in Figure 8. This connection
occurs whenever the relevant world knowledge forms a
prepackaged unit of properties and actions — what Schank and
Abelson (1977) call a script. In such cases the reader may
establish a connection between some node standing for the
input and the entire script. This connection can also benefit
memory retrieval because it allows the reader to encode the
input by constructing a single link to a pre-existent
higher-unit/ and to subsequently retrieve the input by tracing
that link and unpacking the constituents of the unit.
The above account can be made more precise by being more
specific about the world knowledge involved and how it is used
to generate inferences. To aid in this, we need the notion of
a schema y which many have taken to be the basic form of
representation for units of world knowledge (e.g., Adams &
C llins, 1979; Rumelhart & Ortony, 1977). Roughly, a schema
is a description of a particular set of interrelated concepts
that may represent a specific situation, such as going to a
movie, or a general activity that can occur in many
situations, such as asking someone for a favor. The
Organization of Factual Knowledge
53
components of a schema (either other schemas or primitive
concepts) are often only vaguely specified? this permits them
to function as variables that can be filled in or instantiated
by input information with certain properties. To see what
these ideas buy us, we will look at a specific schema and see
how it is used in understanding and remember ing .
2. Schemas f or specific situations ; As developed by
Schank and Abelson (1977) , a script represents the objects and
actions that typically occur in a recurrent, stereotyped
situation. Figure 9 presents a hypothetical script for going
to the movies.
Our script contains several components. First, there^s
the h eader or title, "Going-to-a-Movie," whose major function is
to access the script. Anytime you read something that means
movie-going, you presumably retrieve the script. Second, a
script contains a list of the objects, called props , and of
the roles that are likely to be encountered in a situation
described by the script (see Figure 9). Mention of these
props or roles can also access the script. Third, a script
contains pre-conditions and outcome conditions (see Figure 9) ,
which can again access the script, and also are plausible
inferences given the script has been accessed. For example,
if you read "Herb went to a movie," you can infer Herb had
Organization of Factual Knowledge
54
Header ; Going to a Movie
Props ; Theater, Tickets, Candycounter , Candy, Seats, Film
Roles ; Customer , Cashi er , Re f r eshment Vendor , Usher , Owner
Pre-Condi tions ; Goal of Seeing Movie, Money, Time
Outcome Conditions ; Less Money, Knowledge of Film
Actions ;
Getting Tickets Watching Film
Customer stands ia line Customer Enters Interior
enables enables
Customer Gives Cashier Money Customer Finds Seat
result enables
Cashier Gives Customer Tickets Customer Watches Film
Getting Refreshments
Customer Orders Candy
result
Vendor Gives Customer Ca w 7
result
Customer Gives Vendor Money
Leaving
Customer Leaves Interior
enables
Customer Exits Theater
Figure 9. A sample script for moviegoing.
Organization of Factual Knowledge
55
some money before entering the movie and less of it when he
got out.
Fourth and most importantly / a script consists of the
specific actions likely to occur in the situation. These
actions can be grouped into chunks called scenes (Cullingford,
1978). The script in Figure 9 contains four scenes —
Getting Tickets, Getting Refreshments, Watching Film, and
Leaving — and under each is listed the actions that comprise
it. Note that the props and roles mentioned in the actions
are schema variables; e.g.. Customer or Cashier name
variables that can be filled in by a person playing that role
in the story. Also note that successive actions are
connected by labeled relations; these are critical for
co-nprehension and retrieval processes.
The script actions make up most of the plausible
inferences one can draw when reading a story based on ♦-he
script. To appreciate this, consider how our script r^^n e
us^d to understand and subsequently retrieve the followi.*^;,
vignette.
(1) Herb wanted to see a movie.
(2) He got a cicket.
(3) He found a seat up front.
EKLC
Organization of Factual Knowledge
56
When Sentence (1) is presented, it accesses the
Movie-going script because it mentions a precondition, i.e..
Herb had a goal of seeing a film. Also, Herb will be bound to
the role of Customer. Once the script is accessed, the reader
is expecting something from the Getting Tickets scene. This
expectation is confirmed by Sentence (2) , which matches the
4
script action "Cashier Gives Customer Tickets." At this
point our reader can infer some of the script actions in the
first scene that were not explicitly mentioned. We will
assume that only those actions needed to interrelate the
explicitly mentioned facts are inferred. For example, our
reader might infer that Herb gave the Cashier money, for this
proposition interconnects the first two explicitly mentioned
ones; i.e., wanting to see a movie was the reason for Herb
giving the Cashier money, and the latter resulted in Herb
getting a ticket. Because Sentence (2) marks the end of the
Getting Tickets scene, our reader will now be expecting
something from the Getting Ref r n^^nts scene or, since ^
latter is optional, something from che Watching Film scene.
Sentence (3) matches an action in the Watching Film Scene.
Now our reader can infer some of the actions between the end
of the first scene (explicitly mentioned in the preceding
sentence) and the Find a Seat action of the third scene
(explicitly mentioned in the current sentence) . Again, she
Organization of Factual Knowledge
57
will presumably infer those actions needed to relate the
explicitly mentioned sentences, e.g., she might infer that
Herb entered the interior of the theater because i is
proposition interconnects sentences (2) and (3).
In general, then, one matches each stated fact to a
script action, and one infers nonstated script actions falling
between stated ones that are needed to relate input facts.
The resulting representation for our vignette looks like that
in Figure 10. it contains the input facts, some inferred
script actions, and relations between all propositions. It
also contains a pointer from the node for Herb to the
movie-going script itself.
Consider now two hypotheses about how information in
this representation could be retrieved. In the higher-unit
hypothesis (Smith et al., 1978), our respondent would first
follow the link from Herb to the script itself. If the task
required recall, could read the actions off the script.
We further assume cher; those script actions, corresponding to
(a) stated facts and (b) inferences needed to connect such
facts, are explicitly tagged as such, and that these tags are
used as guides to recall. If the task was one of recognition,
then after accessing the script, she would match each marked
script action to the probe until she found a match. This
Organization of Factual Knowledge
58
Wanted to See Movie
reason for
Gave Cashier Money
result
Got Ticket from Cashier
enable
Entered Interior of Theater
enable
Found Seat
Figure 10- Example of representation for moviegoing
vignette after script processing (see text) •
Organization of Factual Knowledge
59
process seems consistent with the experimental findings on
recall and recognition accuracy. Thus for script-based facts /
recall and recognition accuracy should be relatively high
because only one new link .eed be examined and only one unit
accessed in order to recover all presented facts. But good
performance on the presented items would be purchased at the
cost of an increase in memory confusions because all script
actions corresponding to inferences drawn during comprehension
are candidates for retrieval.
As for the results from the fan experiments^ there should
be little fan effect with integrated facts because the same
higher unit,, the script^ is accessed regardless of how many
facts relevant to the script have been learned. Related
diatractors should be difficult to reject because they often
match tagged script actions that correspond to inferences
drawn during comprehension. Lastly, there is the finding that
latency increases with the number of themes learned about a
character, if each theme corresponds to a script, an increase
in the number of scripts means an increase in the number of
script-links off of the node for the main character (Herb in
the above example) , and *-his fanning will slow down the search
for the relevant higher unit.
Organization of Factual Knowledge
60
Unfortunately^ the higher-unit hypothesis has limited
applicability. For one thing^ the above process seems useless
to someone who read two stories about two different people
engaging in the same script^ for it would confuse the facts
about one person with those about the other . Another
difficulty is that the hypothesis is liiuited to situations for
which people presumably have scripts. This suggests that
facts integratable by scripts will behave differently than
those integratable by other kinds of schemas. The little data
available on this point show no evidence for such a difference
(Reder & Anderson, 1980) .
The second hypothesis is the Interconnections hypothesis.
(It is somewhat similar to Ander song's (1976) , notion of
elaboration.) It ignores the script entirely at the time of
retrieval and op'^rates instead on the interconnected
propositions in the representation. If asked to recall the
story about Herb, our reader would start searching links from
Herb. If she can retrieve any one of the input facts, she
has a direct path to the others since all were interconnected
by inferences. If she cannot retrieve any input fact but can
access an inference made during comprehension, this will get
her to the input since all propositions are interconnected.
Hence script-based facts should be well recalled because of
Organization of Factual Knowledge
61
their intetconnectedness, but at the price of intrusions since
all inferences drawn during comprehension are candidates for
recall. For a recognition task^ the process operates slightly
dif f(-rently . If presented the probe ^ "Herb found a seat up
fronts our reader would first access a stored fact about
Herb, compare it to the probe^ and respond 'bid" if there was a
match. If no match was founds our reader would follow the
connections . from the accessed fact to see if any of them lead
to a proposition that matches the probe. Again/ recognition
of facts actually presented should be relatively accurate
because they are all interconnected^ but at the cost of false
alarms to inferences that are also part of the connected
ne twork.
The interconnectioi>s hypothesis seems to have something
of a problem/ though^ in accounting for the results from the
fan experiments. Specif ically^ while inferences drawn during
comprehension connect input facts^ they do so at the expense
of increasing the fanning off of predicate nodes (i.e.p the
interco-^nections are typically relations between predicates) .
This caused nc difficulty in explaining how integration
facilitated recall or recognition accuracy / because every
link from a predicate node eventually leads to another input
fact/ and this could increase accuracy. But there is a
6*5
Organization of Factual Knowledge
62
difficulty in explaining hov/ an increase in links off the
predicate node can ever facilitate recognition latency,- A
possible solution to this problem is to note that a link
between predicates essentially allows one to access an entire
proposition without going through its terminal nodes • That
is, given the retrieval process has failed to match the probe
to Proposition h, and given an inferential link leading from
predicate A to Predicate one can access Proposition B
without going through the terminal nodes of the probe again.
This facilitation of memory access may more than compensate
for the increase in comparison or search time due to the extra
link off the predicate node.^ The other findings from the
fan experiments cause no further problems for the
interconnections hypothesis. Related distractors are
difficult to reject because they often match inferences made
during comprehension. Finally/ latency increases with the
number of themes learned about a chairacter because there are
no inferential relations between the facts associated with one
theme and those associated with another.
Note that the interconnections hypothesis avoi<5s the
problems that plagued its predecessor. Since the script
itself plays no role in retrieval/ the hypothesis can handle
the situation where one reads and retrieves multiple stories
6-C
Organijjation of Factual Knowledge
63
based on the same script. Getting the script out of the
retrieval process also takes care of another problem? no
longer need there be any major difference between
integration-via-scripts and integration-via-cther-kinds-
of-schemas. According to the present hypothesis^ all scripts
do for memory is interconnect propositions^ and any kind of
schema that can make comparable interconnections should lead
to comparable results.^
3, A comparison of the two kinds of mechanisms . This
has been a long section^ and I had best summarizs the major
issues. To account for integration effects^ we considered
two kinds of theoretical mechanisms • The first assumed that
subdivided networks were a sufficient representation to handle
the effects of integration. The critical processing ideas
were that: (a) since activation of a subnode is sufficient
for recognition of a probe/ the memorized fact corresponding
to the probe need not be retrieved; and (b) some of the
subnode activation was due to concepts that were previously
connected to the subnode and that occurred in the probe. The
second kind of mechanism focused on a different kind of
representation^ namely / a network of interrelated
propositions / 9or;;e corresponding to input facts and others to
inferences. This led to both the higher-unit and
Organization of Factual Knowledge
64
interconnections hypotheses^ but since the former was argued
to be of limited applicability ^ we will consider only the
interconnections hypothesis in what follows •
There are obviously many differences between the two
kinds of mechanisms^ but at a general level the critical
difference seems to be the following ; the interconnections
mechanism focuses on comprehension and assumes memory effects
are consequences of comprehension processes; subnode
activation focuses on memory per se and assumes memory for
integrated facts can be accounted for without a thorough
analysis of how the facts were initially comprehended. Given
this general difference, specific differences fall into place.
Thus in the interconnections hypothesis, we emphasized the
role of world-knowledge inferences because no account of
comprehension can do without them; in the subnode approach,
little or nothing was s id about inferences, not because Reder
and Anderson (1980) do not believe inferences are needed in
comprehension, but because their account of memory phenomena
is not based on comprehension. Then there is the difference
in parsimony. Subnode activation clearly seems the more
parsimonious of the two when it comes to explaining memory
data, but this may be the result of ignoring comprehension.
That is, if Reder and Anderson had to stipulate what is
O
EKLC
Organization of Factual Knowledge
65
involved in comprehending integrated facts^ they might end
up positing representational and processing aspects that look
like those in the schema-based interconnections approach/ and
their edge in parsimony would be gone.
Though the key difference between the approaches is a
general one, there may be a way of bringing some specific data
to bear on a choice between mechanisms. In the subnode
approach/ it seems that activation from any concept connected
to an operative subnode can contribute to recognition? in the
interconnections approach/ only inferences needed for
comprehension can enter into the recognition process. This
contrast can be illustrated by an experiment I recently
7
performed .
Subjects first read four scripts-based stories, each
consisting of seven propositions. For example/ one story was:
Jane went to a restaurant. She went to a table and
sat down. Then she drank a glass of water and ate a
sandwich. Later she paid the claack with cash and
went to get her coat.
Later/ subjects had to decide whether each of a series of
probe sentences "followed" or "did not follow" from one of the
stories. According to the subnode idea/ a subject^'s
Organization of Factual Knowledge
66
representation of the stories would consist of: (a) four
subnodesy one per story (e.g.^ Restaurants) ^ with each being
attached to the seven specific propositions in that story;
and (b) connections between each subnode and all concepts
previously known to be related to that subnode. PreiSumably,
subjects would decide whether or not a probe item follows from
a story partly on the basis of whether or not the concepts
mentioned in the probe activate the prior-knowledge concepts
connected to any subnodes. This predicts that any probe
mentioning a frequent script action should be judged to follow
from that script-based story. But this simply was not the
case. If a probe mentioned a script action that was in no way
needed to understand the original story, subjects uniformly
agreed it did not follow from the story. To illustrate with
the above restaurant story^ the probe^ "Jane ordered dessert/"
was judged by virtually all subjects not to follow from the
story. Yet this probe corresponds to a very frequent action
in the Restaurant script^ more frequent than the script action
corresponding to "Jane got up from the table" (as determined
by the Bower et al. ^ 1979 ^ norms)/ where the latter probe was
judged to follow from the story presumably because it was
needed in understanding.
Organization of Factual Knowledge
67
Having tried to make a case for favoring the
comprehension approach, let us close this section on an
even-handed note by pointing out that even the
interconnections hypothesis must give some role to subdivided
networks. For script-based stories, if a character engages in
activities from two or more unrelated scripts, the final
representation would likely be in the form of a subdivided
network: each branch of the network would contain its own set
of interrelated input facts and inferences, and the subnodes
would be the relevant script headers.
Facts with Correlated Predicates
Empirical Evidence
As best we know, the organizational condition of present
interest has been explicitly studied only in a series of fan
experiments that we recently conducted (Whitlow, Medin, &
Smith, Note 1) .
Fan experiments . In our initial experiment, subjects
learned either one, two, or three facts about a person
designated by an occupation term. Half the subjects learned
facts like those on the left side of Table 4, the other half
learned facts like those on the right side. The only
difference between the two sets of facts is that the
Organization of Factual Knowledge
68
predicates on the left are perfectly correlated whereas those
on the right are not. For the sentences on the left^ if
someone "cleaned the wall/' he also "pushed the trucks" while
if someone "moved the bucket," that^s all he did; not so for
the sentences on the rights where if someone "cleaned the
wall," he might have "pushed the truck" or he might not have.
Since all previously published studies of the fan effect used
less than perfectly correlated predicates^ we wanted to see
if this effect held up when the predicates were perfectly
correlated.
Learning was followed by the usual speeded recognition
task. For correct responses to both Old and New items / we
determined the fan effects separately for perfectly correlated
predicates and for less than perfectly correlated ones. The
results are in the first two rows of Table 5 (magnitude of
the fan effect is estimated by subtracting the latency for the
fan-1 condition from that for the fan-3 condition) . There was
a substantial fan effect when the predicates were less than
perfectly correlated but not when they were perfectly
correlated.
At first we thought our results could be due to the
following. With less than perfectly correlated predicates, a
particular predicate, e.g., "pushed the truck," sometimes
7r>
Table 4
Example of Sentences Used in Whitlow, Medin, and Smith (iJote 1)
Perfectly Correlated Pr_e_dicates Less than Perfectly Correlated Predicates
The banker moved the bucket The banker moved the bucket
The artist moved the bucket The artist cleaned the wall
The lawyer cleaned the wall The lawyer cleaned the wall
The lawyer pushed the truck The lawyer pushed the truck ' '
0
The farmer cleaned the wall The farmer moved the bucket 2
The farmer pushed the truck The farmer pushed the truck j;-
m
It
H-
. 0
3
0
Ml
^ 0
H
%
0
t
H
(D
a
ERIC h
Table 5
Magnitude of Pan Effects in Msec (Fan 3-Fan 1)
for Whitlow, Medin, and Smith Studies
Perfectly Correlated
Predicates
("Old
150
i
Experiment 1
New
■
-100
Concrete \
fold
-140
^New
- 10
Experiment 2
Abstract i
^Old
80
^New
- 75
Experiment 3 i
'Old
-130
^New
- 60
Mean
- 70
Less than Perfectly
Correlated Predicates
250
200
650
110
225
125
540
100
275
Organization of Factual Knowledge
71
occurs in the context of one predicate, "cleaned the wall,"
and sometimes in the context of another, "moved the bucket"
(see Table 4). If context alters meaning, then the meanings
of the less than perfectly correlated predicates were more
variable than those of the perfectly correlated predicates,
and this might have determined whether or not a fan effect
occurred. A second experiment, however, convinced us that
meaning variability was not the critical factor.
Again, one group of subjects learned facts with perfectly
correlated predicates, another learned facts with less than
perfectly correlated predicates, in addition, each group was
split into two subgroups: one worked with concrete predicates
(e.g., "lifted the bucket") and was instructed to think of a
particular predicate the same way when it occurred in
different contexts (i.e., with different companion
predicates) ; the other subgroup worked with abstract
predicates (e.g., "moved the object") and was instructed to
think of a particular predicate in different ways when it
occurred in different contexts. The former subgroup should
have exper ienced less meaning variability. Though this
variation affected recognition latencies, it did not determine
whether or not there was a fan effect. We again found fan
effects only when predicates were less than perfectly
correlated (see the middle rows of Table 5).
Organization of Factual Knowledge
72
A last experiment sought to rule out a possible artifact
(brought to our attention in a person conununication by G.
Bower in 1978) . When subjects learn facts that always have
perfectly correlated predicates^ they could adopt a
task-specific strategy. During learning^ they could tag each
occupation term and each predicate with its fan level (one^
twoy or* three)/ and then during recognition^ they could
respond Old to any probe whose occupation term and predicate
had the same fan level. A glance back at Table 4 should
convince yci: this strategy always yields correct
recognitions only for sentences with perfectly correlated
predicates. Thus in our previous experiments^ this strategy
was available only to subjects who worked with perfectly
correlated predicates and could be the reason they showed no
fan effect.
To discourage this strategy^ we had all subjects learn
two sets of sentences/ half having perfectly correlated
predicates/ and half less than perfectly correlated ones.
With this design/ subjects should be unlikely to use the above
strategy since it would frequently produce incorrect decisions
on facts with less than perfectly correlated predicates. The
results/ presented in the bottom rows of Table 5/ replicated
our previous findings. Apparently the power of perfectly
Organization of Factual Knowledge
73
correlated predicates to offset the fan effect is not due t
the use of a specific strategy.
Implications fo r memory for real-world topics . In the
above^ we described the critical variable as perfectly
correlated vs. less than perfectly correlated predicates .
While literally correct, this is probably misleading.
Since each predicate occurred just twice in our studies, only
a limited range of correlations was possible, and our subjects
may have been able to detect a correlation only among
perfectly correlated predicates. The less than perfectly
correlated predicates may have been perceived as uncorrelated,
and a better description of our variable may be correlated vs.
uncorrelated predicates . Following this line of argument, we
suspect we could substantially reduce the fan effect with any
set of predicates having a noticeable correlation.
If correct, the above conjecture has an important
implication for fan effects. Since predicates about
real-world entities or objects tend to be substantially
correlated, one would expect little change in retrieval
efficiency as more facts are learned about a real-world
entity. To illustrate, consider classes of real-world
objects, like various kinds of animals, plants, and human
artifacts. As Rosch (e.g., 1978) has argued, the predicates
Organization of Factual Knowledge
74
associated with such classes tend to be highly correlated.
Creatures with feathers^ for example^ also have wings and tend
to fly; so if you already know that robins have feathers and
wingS/ then learning that they also fly should not retard the
efficiency of the retrieval process the way learning an
uncor related predicate would. More generally^ to the extent
the world comes in packages of correlated predicates^ there
may be little retrieval interference engendered by learning
multiple facts about the same topic.
Theoretical Mechanisms
In what follows, we briefly consider how well the
theoretical mechanisms already discussed can be extended to
account for the lack of a fan effect with perfectly correlated
predicate^:;. We can start with the subnode-act ivation
hypothesis. When our subjects came across a set of study
sentences where everyone who "cleaned the wall" also "pushed
the truck/" they may have assumed there was a category of
people who clean walls and push trucks. Subjects might then
use these categories as subnodes in a network like that in
Figure 11. Here^ the bottom nodes refer to the people who are
members of each subnode or category. While the representation
seems plausible^ the subnode-act ivation hypothesis cannot
account for the obtained lack of fan effects. Recall that
Organization of Factual Knowlodge
75
this hypothesis explains the lack of fan effects as follows:
a subnode can be activated before any newly learned fact
because the subnode receives activation from concepts that
have been previously linked to it and that also occur in the
probe . In the present case , however ^ any subnode that
corresponds to a set of predicates is novel. Therefore^ there
may not be any concepts previously linked to it 9 which means
there is no way for a subnode to be activated prior to
activation of the learned facts.
AS for the higher-unit hypothesis^ we again assume
subjects treat correlated predicates as defining a category of
people. They would set up a higher-unit for each category
that contains the correlated predicates characterizing it.
Information about an occupation term could be encoded by a
single connection to the appropriate higher-unit. If later
asked to recognize any proposition about a particular
occupation term^ subjects need consider only one link to
access the higher -unit, and could then unpack the unit. So
this hypothesis is consistent with the lack of fan effects.
Fiially^ the interconnections hypothesis would assume
that when presented correlated predicates^ subjects infer a
co-occurrence relation between them^ thereby interconnecting
the input facts. When later required to recognize a study
Organization of Factual Knov/ledge
76
List
\
Bucket- Wall'-clGaning-
mover s tr uck-push er s
\ / \
Banker Artist Lawyer Farmer
Figure 11. Segment of a possible subdivided network for
correlated predicates (see text) .
Organization of Factual Knowledge
77
sentence y subjects would use these inter-predicate connections
the way they presumably use interconnections established when
reading integratable facts. That is^ they would use the
inter-predicate connections as an aid in accessing a new
proposition,
Sununary and Conclusions
Recapitulation of Major Points
Of one thing there is no doubt. When learning multiple
facts about the same topic^ various factors induce us to
organize the material, and this will lead to retrieval of the
input facts that is substantially better than would be
predicted by current models of sentence memory. This boost in
retrieval shows up in three different memory indicators —
recall accuracy, recognition accuracy, and recognition latency
(i.e., reduced fan effects). More specifically: (a) Facts
that subdivide into distinct groups lead to reduced fan
effects and to increased recall accuracy. (b) Facts
integratable by prior knowledge can result in reduced fan
effects, as well as in increases in recall and recognition
accuracy (though at the cost of thematically related
intrusions in recall and poorer performance on related
distractors in recognition). And (c) facts containing
correlated predicates lead to reduced fan effects.
Organization of Factual Knowledge
78
With i-egard to theoretical mechanisms^ things are less
clear* There is consensus on only one point — that a
subdivided network can be used to organize facts from distiact
groups. More precisely: (a) if people learn facts from
distinct groups; and (b) show a distinctive pattern on a
speeded recognition task (increased latency with increases in
the relevant fan but not with increases in the irrelevant
fan)/ or another distinctive pattern on a recall test
f "ome-or-none recall clusters); then (c) they have represented
the input facts in terms of a subdivided network.
When we move to our second organizational condition
facts integratable by prior knowledge — theoretical opinions
diverge. The subnode activation hypothesis holds that the
facts are represented by a subdivided network with themes
serving as the subnodes. As long as most distr actors are
unrelated to the themes, people can use activation of a
subnode as a basis for recognition, part of this activation
coming from concepts that were previously linked to the
subnode and that now appear in the probe. Since specific
facts need not be accessed, retrieval is rapid and independent
of the number of facts learned about a topic. The alternative
position focuses on the comprehension of integratable facts.
It holds that comprehension involves using schemas to draw
Organizcition of Factual Knowledge
79
inferences about the input, as well as possibly establishing a
link between some component of the input and the schema
itself. Some inferences interconnect the input facts,
thereby providing alternate access routes during recall or
recognition, which boost memory in both kinds of tasks. If a
connection between the input and the schema itself has been
established, retrieval can be accomplished by simply accessing
the schema and reading the input facts off of it. As only one
link need be accessed, retrieval should be rapid and
independent of the number of schema-based facts that were
lear ned.
When we tried to extend these hypotheses to explain the
lack of fan effects with correlated predicates, the clearest
result was that the subnode activation hypothesis failed.
Specifically, while it seemed plausible that each set of
correlated predicates was dominated by a single subnode in a
subdivided network, such a subnode was likely a novel concept
and hence unlikely to have concepts previously linked to it
that could contribute to its activation.
While the above hardly provides firm answers to the
theoretical questions raised in the Introduction, it does
suggest no one mechanism is going to cover all organizational
conditioHL^ Thus, subd ivided networks seem our best
Organization of Factual Knowledge
80
contender for describing what goes on with facts from distinct
groups^ but an unlikely alternative for explaining the results
for facts containing correlated predicates. Similarly^ our
interconnections hypothesis works best in explaining
variations in memory accuracy with integratable facts^ and
seems beside the point when it comes to accounting for results
with facts from distinct groups. So we may need all three
mechanisms — subdivided networks^ higher units, and
interconnections .
The Status of the Fan Effect
One strategy followed throughout is to take the fan
effect as a kind of landmark, and to use reductions of this
effect as indicators^ of organizational factors. While the
theoretical importance of the fan effect seems to justify
this strategy, some comment is in order about the limited
generality of this effect.
We have seen that any one of three different factors can
reduce the fan effect. Hence, a substantial fan effect occurs
only when the facts to be learned conform to the following
conjunction of negative conditions: (a) the facts are not
from distinct groups, (b) they are not readily integratable by
prior knowledge, and (c) they do not contain correlated
Organization of Factual Knowledge
81
predicates. The work of Hayes-Roth (1977) supplies still
another negative factor: (d) the facts are not well practiced.
So a fan effect is obtainable in the laboratory only under a
choice of parameters that captures a four-fold conjunction of
negative conditions. This means the effect is not among our
most robust laboratory phenomena. Furthermore, the above
conjunction of negative conditions may rarely occur in real
life. The vast majority of real-life learning situations
involve facts that are integratable by prior knowledge and/or
have correlated predicates. Most times that we read text (or
listen to utterances) , we are exposed to multiple facts
about a topic that are integratable by prior knowledge; if
this was not the case, we would probably judge the text
incoherent. And when we think of real-life cases where the
facts presented are not integratable by prior knowledge, the
situations that come most readily to mind are where we learn a
novel concept. Here, the predicates of the facts are often
highly correlated.
The weak point in the preceding is that we are using
laboratory experiments with a very restricted variation o;
fanning (generally from one to three) to draw implications
about real-life situations that may have a far greater
variation of fanning. Thus many real-life situations may have
Organization of Factual Knowledge
82
a fanning variation of 1 to 100 (e.g.^ How much do you know
about the Mayor of San Francisco vs. about the President of
America?) , and this huge variation may result in a substantial
fan effect even in situations where our conjunction of
negative conditions does not hold. The only way to check this
is to perform laboratory experiments with conditions known to
reduce the fan effect but with huge variations in fanning.
Without such experiments^ we run the risk of studying a
phenomenon that rarely occurs outside of the laboratory.
Even if such experiments are performed and do yield
substantial fan effects^ there is still a problem in focusing
so much effort on laboratory situations defined by the above
conjunction of negative conditions. For the representations
and processes operative in situations that do not meet the
conjunction of negative conditions may be qualitatively
different from those operative in situations that do meet our
conjunction. We saw a good example of this in the fan
experiments dealing with integratable facts (at least in those
using unrelated distractors) . Even Reder and Anderson^'s
(1980) account of these results introduced some new
representational aspects — namely^ the subnodes — and new
processing assumptions — namely^ that activation of a subnode
could trigger a recognition decision. These new aspects are
^8
Organization of Factual Knowledge
83
qualitatively different from the entities in Anderson^'s (1976;
Chapter 8) ACT theory of sentence memory (thou'^^ih readily
interfaceable with that theory) , and we would be unlikely to
think of these new aspects unless people did research on
paradigms that are not specifically configured to yield fan
effects.
A Comprehension Approach to Memory Phenomena
In discussing organizational mechanisms for integrated
facts, we argued that comprehension processes ^ like
inferencing^ may lie behind memory effects. Essentially ^ we
singled out facts integratable by prior knowledge as the one
condition where we need to consider comprehension in order to
understand memory. This argument can readily be extended.
Namely^ whenever we deal with memory for facts about the same
topic y we first need to understand what goes on in the
comprehension of these facts ,
Let us go back to the beginning. We started by
considering facts from distinct groups. No mention was made
there of comprehension. Instead^ we noted that the memory
representation for such facts often consists of a subdivided
network^ and traced the implications of this for retrieval.
But, why is such a representation constructed? One
Organization of Factual Knowledge
84
possibility is that it facilitates retrieval. Another is that
a subdivided network is the natural consequence of our
comprehension processes operating on an input where the most
salient relations between the facts are that some belong to
one group^ while others belong to different groups. That is^
if the business of comprehension processes is to find
relations between input facts^ and the only salient relation
is that some facts are members of the class of statements
about countries while others are members of the class of
statements about animals, then all the comprehension processes
can do is construct a representation that depicts these
class-membership relations. In short, subdivided networks are
a kind of representation you get out of comprehension
processes when your input is sparse on relations.
As for the studies involving facts with correlated
predicates, we again have a case where the input is sparse on
relations. The only relation the comprehension processes can
pick up on here is that some predicates co-occur with others.
To sum up, we may have underestimated the extent to which
memory phenomena are dependent on comprehension by
consistently using materials that lack the stuff that makes
comprehension go — relations. Research concerned with memory
for integratable facts may be the only way to redress this
imbalance .
Organization of Factual Knowledge
85
Reference Notes
Whitlow, Vi.J., Medin, D.L., & Smith, E.E. Retrieval of
correlated predicates . Unpublished manuscript, Rutgers
University, 1980.
McClosky, ^5. Search and c omparisons processes in fact
retrieval and question answering . Unpublished manuscript,
Johns Hopkins University, 1979.
Organization of Factual Knowledge
86
References
Adams/ M.J.^ & Collins^ A.M. A schema-theoretic view of
reading. in R.O. Preedle (Ed.), New directions in
discourse processing . Norwood, N.J.: Ablex, 1979.
Anderson, J.R. Retrieval of prepositional information from
long-term memory. Cognitive Psychology , 1974, 5, 451-474.
Anderson, J.R. Item-specific and relation-specific
interference in sentence memory. Journal of Experimen tal
Psychology ; Human Le ar n ing and Memory , 1975, 104,
249-260.
Anderson, J.R. Language, memory, and thought . Hillsdale,
N.J.: Erlbaum, 1976.
Anderson, J.R., & Bower, G.H. Human associative memory .
Washington, D.C.: Winston, 1973.
Anderson, J.R., & Paulson, R. Interference in memory for
pictorial information. Cogni t ive Psychology , 1978, 10,
178-202.
Black, J.B., & Bower, G.H. Episodes as chunks in narrative
memory. Journal of Verbal Learning and Verbal Behavi or ,
1980, in press.
£'5
Organization of Factual Knowledge
87
Bousfield, Wo A. The occurrence of clustering in the recall of
randomly arranged associates . Journal of General
Psychology , 1953, 36, 67-81.
Bousfield, W.A., & Cohen, B.H. The effects of reinforcement
on the occurrence ^f clustering in the recall of randomly
arranged associates. Jou r nal of Psychology , 1953, 36 ,
67-81.
Bower, G.H., Black, J.B., & Turner, T.J. Scripts in memory
for text. C ognitive Psycholo9y , 1979, 11, 177-220.
Bower, G.H., Clark, M. , Winzenz, D. , & Lesgold, A.
Hierarchical retrieval schemes in recall of categorized
word lists. Journal of Verbal Learning and Verbal
Behavior , 1969, 8, 323-343.
Bransford^ J.D. , & Johnson, M.K. Considerations of some
problems of comprehension. In W.G. Chase (Ed.), Visual
information processing . New York: Academic Press, 1973.
Bransford, J.D. , & McCarrell, N.S. A sketch of a cognitive
approach to comprehension: Some thoughts about what it
means to comprehend. In W.B. Weimer & D.S. Palermo
(Eds.), Cognition and symbolic processes . New
York : Winston, 1975 .
EKLC
9o
Organization of Factual Knowledge
88
Cofer, C.N., Bruce, D.R., & Reicher, G.M. Clustering in free
recall as a function of certain methodological variations.
Journal of Experimental Psycho logy , 1966, 71, 858-866.
Cohen^ B.H. Recall of categorized word lists. Journal of
Experimental Psychology , 1963, 66, 227-234.
Cullingford, R.E. Script application; Computer understanding
of newspaper stories. (Research Report No. 116) . New
Haven, Conn.: Yale University, Department of Computer
Science, 1978.
Dooling, D. J. , & Lachman, R. Effects of comprehension on
retention of prose. Journal of Experimental Psychology ,
1971, 8i, 216-222.
Hayes-Roth, B. Evolution of cognitive structure and
processes. Ps ychological Revi ew, 1977, 84, 260-278.
Hayes-Roth, B., & Hayes-Roth, F. The prominence of lexical
information in memory representations of meaning. Journal
of Verba l Learning and Verbal Behavior , 1977, 16, 119-136.
King, D.R.^ & Anderson, J.R. Long-term memory search: An
intersecting activation prooes:*^. Journal of Verbal
Learning and Verbal Behavior , 1976, 15^, 587-606.
EKLC
9i
Organization of Factual Knowledge
89
Kolodner, J.L. Memory organization for natural language
data-base base inquiry. (Research Report No. 142). New
Haven, Conn.: Yale University, Department of Computer
Science, 1978.
Lewis, C.H., & Anderson, J.R. Interference with real world
knowledge. Cognitive Psychology , 1976, 8, 311-335.
Handler, G. Organization and memory, in K.W. Spence & j.T.
Spence (Eds.), The psychology of learning and motivation
(Vol, 1). New York: Academic Press, 1967.
Minsky, M. A framework for representing knowledge. m p.H.
Winston (Ed.), The sychology of computer vision . New
York: McGraw-Hill, 1975.
Moesher, S.D. The role of experimental design in
investigation of the fan effect. Journal of Exper im ental
Psychology: Human Learning and Memory , 1979, 5, 125-134.
Nelson, T.O., & Smith, E.E. Acquisition and forgetting of
hierarchically organized information in long-term memory.
Journal of Experimental Psychology , 1972, 95, 388-396.
Norman, D.A., & Bobrow, D.G. On the role of active memory
processes in perception and cognition. In C.N. Cofer
(Ed.), The structure of human memory . San
Francisco: Freeman, 1976.
^5
Organization of Factual Knowledge
90
Norman^ D.A.^ & Rumelhart^ D.Eo Explorations in cognition ,
San Francisco: Freeman, 1975.
Puff/ C.R. Role of clustering in free recall. Journal of
Experimental Psychology , 1970, 86, 384-386.
Reder, L.M., & Anderson, J.R. A partial resolution of the
paradox of interference: The role of integrating
knowledge. Cognitive Psychology , 1980, in press.
Rosch, E. Principles of categorization. In E. Rosch & B.B.
Lloyd (Eds.), Cognition and categorization . Potomac,
Md. : Erlbaum, 1979-
Rumelhart, D. Notes on a schema for stories. In D. Bobrow &
A.M. Collins (Eds.), Representations and
understanding: Studies in cognitive sciences . New
York: Picademic Press, 1975.
Rumelhart, D.E., Lindsay, P.H., & Norman, D.A. A process
model for long-term memory. In E. Tulving & W. Donaldson
(Eds. ) , Organization and memory . New York: Academic
Press, 1972.
Rumelhart, D^E., & Ortony, A.. The representation of knowledge
in memory. In R.C. Anderson, R.J. Spiro, & W.E. Montague
(Eds.), Schooling and the acquisition of knowledge .
Hillsdale, N.J.: Erlbaum, 1977.
Organization of Factual Knowledge
91
Rundus, D. Negative effects of using list items as recall
cues. Journal of V erbal Learning and Verbal Behavior ,
1973, 12, 43-50.
Schank, R.C. , & Abelson, R.P. Scripts, plans, goals, and
understanding . Hillsdale, N.J.: Erlbaum, 1977.
Shiffrin, R.M. Memory search. In D.A. Norman (Ed.), Models
of human memory . New York: Academic Press, 1970.
Shoben, E.J., Wescourt, K. , & Smith, E.E. Differential
performance in sentence-verification and
sentence-recognition: Implications for the distinction
between episodic and semantic memory., Journal of
Experimental Psychology; Human Learning and Memory . 1978,
4, ,,,30 4- 3 17.
Smith, E.E. , Adarcs, U. , & Schorr, D. Fact retrieval and th .
paradox of interference. Cognitive Psychology, 1978, 10,
438-464.
Thorndyke, P., & Sower, G.H. Storage and retrieval processes
in sentence memory. Cognitive Psychology , 1974, 5^,
515-543.
Tulving, E. Cue-dependent forgetting. American Scientist ,
1974, 62, 74-82.
Organization of Factual Knowledge
92
Tulving, E. ^ & Pearlstone^ Z. Availability versus
accessibility of information in memory for words. Journal
of Verbal Learning and Verbal Behavior ^ 1966, 5, 381-391.
Winograd, T. Understanding natural language. Cognitive
Psychology , 1972, 3, 1-191.
98
Organization of Factual Knowledge
93
Footnotes
^This prediction was not e>:plicitly made by Rumelhart,
Lindsay^ and Norman (1972) or in any other paper on the ELINOR
model that I know of. However^ I believe it follows quite
directly from what has been explicitly stated about the
model's representations and retrieval processes.
2
This brief review of conditions that foster organization
has omitted Hayes-Roth's (1977) work, which indicates that
practice can organize the constituents of a proposition into a
single unit. The reason for the omission is that Hayes-Roth
focuses on the organization of a single proposition, while I
am concerned with organizing a set of propositions.
3
A substantial increase would occur if there were
numerous facts stored about Diane Keaton.
4
More precisely, the proposition in Sentence (2) matches
a simple inference drawn from the script action "Cashier Gives
Customer Tickets."
^The view of retrieval embodied in this solution makes a
sharp distinction between gaining access to a meinorized
proposition (an access stage) and inspecting the contents of
that proposition (a comparison or search stage) . The proposed
solution assumes that the speed-up in the access stage is
greater than the slow-down in the comparison stage.
99
Organization of Factual Knowledge
94
In particular^ schemata used to encode knowledge about
goals and plans (eJg.^ Rumelhart^ 1975; Schank & Abelson,
1977) should lead to the same kind of memory results as
scripts do.
7
This study was done in collaboration with Mark Chambers
and John Greeno.
ERIC
CENTER FOR THE STUDY OF READING
READING EDUCATION REPORTS
No. 1: Durkin, D. Comprehension Instruction -Where are Voo?, October 1977. (ERIC Document Repro-
duction Service N&. ED 146 566, 14p.. PC-$1.82. IVIF $.83)
No. 2: Asher, S. R. Sex Differences in Reading Achievement, October 1977. (ERIC Document Reproduc-
tion Service No. ED 145 567, 30p.. PC-$3.32. MF-$.83)
No. 3: Adams. M. J., Anderson. R. C, & Durkin, D. Beginning Reading: Theory and Practice, November
1977. (ERIC Document Reproduction Service No. ED 151 722, 15p.. PC-$1.82. MF-$.83)
No. 4: Jenkins. J. R.. & Pany, D. Teaching Reading Comprehension in the Middle Grades, January 1978.
(ERIC Document Reproduction Service No. ED 151 756, 36p., PC-$3.32, MF-$.83)
No. 5: Bruce, B. What Makes a Good Story?, June 1978. (ERIC Document Reproduction Service No.
ED 158 222. 16p.. PC-$1.82, MF-$.83)
No. 6: Anderson, T. H. Another UxJk at the Seif Questioning Study Technique, September 1978. (ERIC
Document Reproduction Service No. ED 163 44i, 19p.. PC-$1.82, MF-$.83)
No. 7: Pearson, P. D., & Kamil. M. L Basic Processes and Instructional Practices in Teaching Reading,
December 1978. (ERIC Document Reproduction Service No. ED 165 118. 29p.. PC-$3.32. MF-$.83)
No. 8: Collins, A. & Haviland. S. E. Children's Reading Problems, June 1979. (ERIC Document Reproduc-
tion Service No. ED 172 188, 19p., PC-$1.82. MF-$.83)
No. 9. Schallert. D. L. & Kleiman, G. M. Some Reasons Why Teachers are Easier to Understand than
Texttnoki, June 1979. (ERIC Document Reproduction Service No. ED 172 189. 17p.. PC-$1.82 MF-
$.83)
No. 10: Baker, L Do I Understand or Do I not Understand: That is thi Question, July 1979. (ERIC
Document Reproduction Service No. ED 174 948, 27p.. PC-$3.32, MF-$.83)
No. 11: Anderson. R. C, & Freebody, P. Vocabulary Kno^vledge and Reading. August 1970. (ERIC Docu-
ment Reproduction Service No. ED 177 470, 52p., PC-$4.82. MF-$.83)
No. 12: Joag-dev. C. & Steffensen. M. S. Studies of the Bicultural Reader: Implications for Teachers and
Librarians, January 1980. (ERIC Document Reproduction Service No. ED 181 430. 28p.. PC-$3 32 MF-
$.83)
No. 13: Adams, M., & Bruce. B. Background Knowledge and Reading Comprehension, January 1980.
(ERIC Document Reproduction Service No. ED 181 431. 48p., PC-$3.32, MF-$.83)
No. 14: Rubin, A. Making Stories, Making Sense (includes a response by T. Raphael and J. LaZansky).
January 1980. (ERIC Document Reproduction Service No. ED 181 432, 42p.. PC-$3.32, MF-$.83)
No. 15: Tierney. R. J., & LaZansky, J. The Rights and Responsibilities of Readers and Writers: A Contrac-
tual Agreement (includes responses by R. N. Kantor and B. B. Armbruster), January 1980. (ERIC Docu-
ment Reproduction Service No. ED 181 447, 32p.. PC-$3.32. MF-$.83)
No. 16: Anderson, T. H.. Armbruster. B. B., & Kantor. R. N. How Qearty Written are Children's Textbooks?
Or, Of Bladderworts and Ma (includes a response by M. Kane, Senior Editor, Ginn and Company),
August 1980.
No. 17: Tierney. R. J., Mosenthal, J., & Kantor, R. N. Some Oassroom Applications of Text Analysis:
Toward Improving Text Selection and Use, August 1980.
10^
CENTER FOR THE STUDY OF READING
. TECHNICAL REPORTS
No. 1: Halff. H. M. Graphical Evaluation of Hierarchical Clustering Schemes. October 1975. (ERiC Docu
ment Reproduction Service- No. ED 134 926. lip.. PC-$1.82. MF-$.83)
No. 2: Spiro. R. J. Inferential Reconstruction in Memory for Connected Discourse. October 1975. (ERIC
Document Reproduction Service No. ED 136 187. 81p.. PC-$6.32. MF $.83)
No. 3: Goetz. E. T. Sentences in Usts and in Connected Discourse. November 1975. (ERIC Document
Reproduction Service No. ED 134 927. 75p.. PC-$4.82. MF-$.83)
No. 4: Alessi. S. M.. Anderson. T. H.. & Biddle. W. B. hia -dyvare and Software Considerations in Computer
Based Course Managament November 1975. (ERIC Document Reproduction Service No ED 134 928
21p.. PC-$1.82. MF $.83)
No. 5: Schallert. D. L. Improving Memory for Prose: The Relationship between Depth of Processing and
Context. November 1975. (ERIC Document Reproduction Service No. ED 134 929. 37p.. PC-$3 32 MF-
$.83)
No S: Anderson. R. C. Goetz. E. T.. Pichert. J. W.. & Kalff. H. M. Two Faces of the Conceptual Peg
Hypothesis, January 1976. (ERIC Document Reproduction Service No. ED 134 930 29p PC-$3 32
MF-$.83)
No. 7: Ortony. A. Names. Descriptions, and Pragmatics. February 1976. (ERIC Document Reproduction
Service No. ED 134 931. 25p.. PC $1.82. MF $.83)
No. 8: Mason. J. M. QuestiorUng the Notion of Independent Processing Stages in Reading. February
1976. [Journal of Educational Psychology. 1977. 69. 288-297)
No. 9: Siegel, M. A. Teacher Behaviors and Curriculum Packages: Implications for Research and
Teacher Education. April 1976. (ERIC Document Reproduction Service No. ED 134 932 42p PC-
$3.32. MF-$.83)
No. 10: Anderson. R. C. Pichert. J. W.. Goetz. E. T.. Schallert. D. L. Stevens. K. C. & Trollip. S. R. Instantia-
tion of General Terms. March 1976. (ERIC Document Reproduction Service No. ED 134 933 30p PC-
$3.32. MF-$.83)
No. 11: Armbruster. 8. B. Learning Principles from Prose: A Cognitive Approach Based on Schema
Theory. July 1976. (ERIC Document Reproduction Service No. ED 134 934. 48p.. PC-$3.32. MF-$.83)
No. 12: Anderson. R. C. Reynolds. R. E.. Schallert. D. L. & Goetz. E. T. Framevmrks for Comprehending
Discourse. July 1976. (ERIC Document Reproduction Service No. ED 134 935. 33p.. PC-$3.32. MF-$.83)
No. 13: Rubin. A D.. Bruce. B. C. & Brown. Jt S. A Process-Oriented Language for Describing Aspects of
Reading Comprehension. November 1976. (ERIC Document Reproduction Service No. ED 136 188
41 p.. PC-$3.32. MF-$.83)
No. 14: Pichert. J. W.. & Anderson. R. C. Taking Different Perspectives on a Story. November 1976.
(ERIC Document Reproduction Service No. ED 134 936. 30p.. PC-$3.32. MF-$.83)
No. 15: Schwartz. R. M. Strategic Processes in Beginning Reading November 1976. (ERIC Document
Reproduction Service No. ED 134 937. 19p.. PC-$1.82. MF-$.83)
No. 16: Jenkins, J. R.. & Pany. D. Curriculum Biases in Reading Achievement Tests. November 1976.
(ERIC Document Reproducticr Service No. ED 134 938. 24p., PC-$1.82. MF-$.83)
No. 17: Asher. S. R.. Hymei. S.. & Wigfield. A. Children's Comprehension of High- and Low-Interest
Material and a Comparison of Two Cloze Scoring Methods. November 1976. (ERIC Document Repro-
duction Service No. ED 134 939. 32p.. PC-$3.32. MF-$.83)
No. 18: Brown. A. L. Smiley, s. S.. Day. J. D.. Townsend. M. A. R.. & Lawton, S. C. Intrusion of a Thematic
Idea in Children's Comprehension and Retention of Stories. December 1976. (ERIC Document Repro-
duction Service No. ED 136 189. 39p.. PC-$3.32. MF-$.83)
No. 19: Kleiman. G. M. 77»e Prelinguistie Cognitive Ba^s of Children's Communicative Intentions. Febru-
ary 1977. (ERIC Document Reproduction Service No. ED 134 940, 51p.. PC-$4.82. MF-$.83)
No. 20: Kleiman. G. M. 77» Effect of Previous Context on Reading Individual Words. February 1977.
(ERIC Document Reproduction Service No. ED 134 941, 76p.. PC-$6.32. MF-$.83)
No. 21: Kane. J H.. & Anderson. R. C Depth of Processing and Interference Effects in the Learning and
Remembering of Sentences. February 1977. (ERIC Document Reproduction Service Wo. ED 134 942
29p.. PC-$3.32. MF-$.83)
102
No. 22: Brown. A. L. & Campione. J. C. Memory Strategies in Ijearning: Training Children to Study Stra-
tegically March 1977. (ERIC Document Reproduction Service No. ED 136 234. 54p.. PC-$4.82. IVIF-
$.83)
No. 23: Smiley. S. S.. Oakley. D. D.. Worthen. D.. Campione. J. C. & Brown. A. L Recall of Thematically
Relevant Material by Adolescent Good and Poor Readers as a Function of Written Versus Oral
Presentation. March 1977. (ERIC Document Reproduction Service No. ED 136 235. 23p.. rC-$1.82,
MF$-.83)
No. 24: Anderson. P C . lipiro. R. J.. & Anderson. M. C. Schemata as Scaffolding for the Representation
of Informatior iii Connected Discourse, March 1977. (ERIC Document Reproduction Service No.
ED 136 236. If if. PC$1.82. MF-$.83)
No. 25: Pany. D.. .• sjns. J. R. Learning Wbrd Meanings: A Comparison of Instruction^ Procedures
ajvd Effects o: ^^.aasures of Reading Comprehension vfith Learning Disabled Students, March 1977.
(ERIC Document Reproduction Service No. ED 136 237. 34p.. PC-$3.32. MF $.83)
No. 26: Armbruster. B. B.. Stevens. R. J.. & Rosenshine. B. Analyzing Content Coverage and Emphasis: A
Study of Three Curricula and Two Tests. March 1977. (ERIC Document Reproduction Service No.
ED 136 238. 22p.. PC-$1.82. MF-$.83)
No. 27: Ortony. A. Reynolds. R. E.. & Arter. J. A Metaphor: Theoretical and Empirical Research, March
1977. (ERIC Document Reproduction Service No. ED 137 752. 63p.. PC-$4.82. MF-$.83)
No. 28: Ortony. A Remembering and Understanding Jabberwocky and Small Talk. March 1977. (ERIC
Document Reproduction Service No. ED 137 753. 36p.. PC-$3.32, MF-$.83)
No. 25: Schallert. D. L. Kleiman. G. M.. & Rubin. A D. Analyses of Differences between Written and Oral
language. April 1977. (ERIC Document Reproduction Service No. ED 144 038. 33p.. PC-$3.32. MF-
$.83)
No. 30: Goetz. E. T., & Osborn. J. Procedures for Sampling Texts and Tasks in Kindergarten through
Eighth Grade. April 1977. (ERIC Document Reproduction Service No. ED 146 565. 80p.. PC-$6.32. MF-
$.83)
No. 31: Nash-Webber. B. Anaphora: A Cross-Disciplinary Survey April 1977. (ERIC Document Reproduc-
tion Service No. ED 144 039. 43p.. PC $3.32. MF $.83)
No. 32: Adams. M. J.. & Collins. A. A Schema-Theoretic View of Reading Comprehension. April 1977.
(ERIC Document Reproduction Service No. ED 142 971. 49p.. PC-$3.32. MF-$.83)
No. 33: Muggins. A. W. F. Syntactic Aspects of Reading Comprehension. April 1977. (ERIC Document
Reproduction Service No. ED 142 972. 68p.. PC $4.82. MF $.83)
No. 34: Bruce. B. C. Plans and Social Actions. April 1977. (ERIC Document Reproduction Service No.
ED 149 328. 45p.. PC $3.3Z MF-$.83)
No. 35: Rubin. AD A Theoretical Taxonomy of the Differences bet¥¥een Oral and Written Language.
January 1978. (ERIC Document Reproduction Service No. ED 150 550. 61p.. PC-$4.82. MF-$.83)
No. 36: Nash-Webber. B.. & Reiter. R. Anaphora and Logical Form: On Formal Meaning Representation
for Natural Language. April 1977. (ERIC Document Reproduction Service No. ED 142 973. 42p.. PC-
$3.32. MF $.83)
No. 37: Adams. M. J. Failures to Comprehend and Levels of Processing in Reading, April 1977. (ERIC
Document Reproduction Service No. ED 145 410, 51p.. PC-$4.82. MF-$.83)
No. 38: Woods. W. A. Multiph Theory Formation in High-Level Perception. April 1977. (ERIC Document
Reproduction Service No. ED 144 020. 58p.. PC-$4.82. MF-$.83)
No. 40: Collins, A. Brown, J. S.. & Larkin, K. M. Inference in Text Understanding. December 1977. (ERIC
Document Reproduction Service No. ED 150 547, 48p.. PC-$3.32, MF $.83)
No. 41: Anderson. R. C. & Pichert J. W. Recall of Previously Unrecallable Information Following a Shift
in Perspective. April 1977. (ERIC Document Reproduction Service No. ED 142 974. 37p.. PC-$3.32.
MF-$.83)
No. 42: Mason, J.. Osborn. J.. & Rosenshine. B. A Consideration of Skill Hierarchy Approaches to the
Teaching of Reading, December 1977. (ERIC Document Reproduction Service No. ED 150 549. 176p..
PC-$1232. MF $.83)
No. 43: Collins. A.. Brown, A. L, Morgan, J. L. & Brewer, W. F. The Analysis of Reading Tasks and Texts.
April 1977. (ERIC Document Reproduction Service No. ED 145 404. 96p.. PC $6.32. MF-$.83)
Nc. 44: McQure. E. Aspects of Code-Switching in tt)e Discourse of Bilingual Mexican American Chihtren.
April 1977. (ERIC Document Reproduction Service No. ED 142 975. 38p., PC $3.3Z MF $.83)
No. 45: Schwartz, R. M. Relation of Context Utilization and Orthographic Automaticity in Word Identifi-
cation, May 1977. (ERIC Document Reproduction Service No. ED 137 762, 27p.. PC-$3.32. MF-$.83)
No. 46: Anderson. R. C. Stevens. K. C, Shifrin, Z.. & Osborn. J. Instantiation of Word Meanings in Chil-
dren, May 1977. (ERIC Document Reproduction Service No. ED 142 976. 22p.. PC $1.82, MF $.83)
No. 47: Brown. A. L Knoviing When, Where, and How to Remember: A Problem of Metacognition, June
1977. (ERIC Document Reproduction Service Wo ED 146 562. 152p.. PC $10.82. MF-$.83)
No. 48: Brown. A. L. & DeLoache. J. S. Skills, Plans, and Self-Regulation, July 1977. (ERIC Document
Reproduction Service No. EO 144 (D40. 66p.. PC-$4.82, MF-$.83)
No. 49: Goetz. E. T. Inferences in the Comprehension of and Memory for Text, July 1977. (ERIC Docu-
ment Reproduction Service No. ED 150 54S. 97p.. PC $6.32. MF-$.83)
No. 50: Anderson. R. C. Schema-Directed Processes in Language Comprehension, July 1977. (ERIC
Document Reproduction Service No. ED 142 S77. 33p.. PC-$3.32. MF-$.83)
No. 51: Brown. A. L Theories of Memory and the Problems of Development: Activity, Growth, and
Knowledge, July 1977. (ERIC Document Reproduction Service No. ED 144 041. 59p. PC $482. MF-
$.83)
N'j. 52: Morgan. J. L Two Types of Convention in Indirect Speech Acts, July 1977. (ERIC Document
Reproduction Service No. ED 145 405, 40p.. PC-$3.32. MF-$.83)
No. 53: Brown. A. L, Smiley. S. S.. & Lawton. S. C. 77»e Effects of Experience on the Selection of Suitable
Retrieval Cues for Studying from Prose Passages, July 1977. (ERIC Document Reproduction Service
No. ED 144 042, 30p.. PC-$3.32. MF-$.83)
No. 54: Fleisher. L S., & Jenkins. J. R. Effects of Contextualized and Decontextualized Practice Condi-
tions on Word Recognition, July 1977: (ERIC Document Reproduction Service No. ED 144 043. 37p..
PC $3.32. MF-$.83)
No. 55: Jenkins, J. R.. & Larson. K. Evaluating Error Correction Procedures for Oral Reading June 1978.
(ERIC Document Reproduction Service No. ED 158 224. 34p.. PC-$3.32. MF-$.83)
No. 56: Anderson. T. H., Standiford. S. N., & Alessi. S. M. Computer Assisted Problem Solving in an Intro-
ductory Statistics Course, August 1977. (ERIC Document Reproduction Service No. ED 146 563. 26p..
Pt:-$3.32. MF-$.83)
No. 57: Barnitz, J. Interreletionship of Orthography and Phonological Structure in Leaning to Read,
August 1977. (ERIC Document Reproduction Service No. ED 150 546. 62p., PC-$4.82. MF-$l83)
No. 58: Mason. J. M. The Role of Strategy in Reading in the Mentally Retarded, September 1977. (ERIC
Document Reproduction Service No. ED 145 406. 28p.. PC-$3.32. MF-$.83)
No. 59: Mason. J. M. Reading Readiness: A Definition and Skills Hierarchy from Preschoolers' Develop-
ing Conceptions of Print September 1977. (ERIC Document Reproduction Service No. ED 145 403.
57p.. PC-$4.82. MF-$.83)
No. 60: Spiro. R. J.. & Esposito. J. J. Superficial Processing of Explicit Inferences in Text, December
1977. (ERIC Document Reproduction Service No. ED 150 545. 27p., PC-$3.32. MF-$.83)
No. 65: Brewer. W. F. Memory for the Pragmatic Implications of Sentences, October 1977. (ERIC Docu-
ment Reproduction Service No. ED 146 5G4. 27p.. PC-$3.32. MF-$.83)
No. 66: Brown. A. L. & Smiley. S. S. 77ie Development of Strata^ for Study Prose Passages, October
1977. (ERIC Document Reproduction Service No. ED 145 371, 59p.. PC-$4.82. MF-$.83)
No. 68: Stein. N. L. & Nezworski. T. 77»e Effects of Organization and Instructional Set on Story Memory,
January 1978. (ERIC Document Reproduction Service No. ED 149 327. 41p.. PC $3.32. MF-$.83)
No. 69: Stein. N. L How Chikiren Understand Stories: A Developmental Analysis, March 1978. (ERIC
Document Reproduction Service No. ED 153 205, 68p.. PC-$4.82. MF $.83)
No. 76: Thieman. T. J.. & Brown. A. L The Effects of Semantic and Formal Similarity on Recognition
Memory for Sentences in CNUren, November 1977. (ERIC Document Reproduction Service No.
ED 150 551. 26p.. PC-$3.32. MF-$.83)
No. 77: Nash-Webber. B. L Inferences in an Approach to Discourse Anaphora, January 1978. (ERIC
Document Reproduction Service No. ED 150 552. 30p.. PC-$3.32. MF-$.83)
No. 78: Gentner. D. On Relation^ Meaning: The Acquisition of Verb Meaning, December 1977. (ERIC
Document Reproduction Service No. ED 149 325. 46p.. PC-$3.32. MF-$.83)
No. 79: Royer. J. M. Theories of iJaarrUng Transfer, January 1978. (ERIC Document Reproduction Service
No. ED 149 326. 55p.. PC $4.82. MF-$.83)
No. 80: Arter. J. A.. & Jenkins. J. R. Differential Diagnosis-Prescriptive Teaching: A Critical Appraisal,
January 1978. (ERIC Document reproduction Service No ED 150 578. 104p.. PC-$7.82. MF-$.83)
No. 81: Shoben. E. J. Choosing a Model of Sentence Picture Comparisons: A Reply to CatUn and Jones,
February 1978. (ERIC Document Reproduction Service No. ED 150 577. 30p.. PC-$3.32. MF-$.83)
104
No. 82: Steffensen, M. S. Bereiter iind Engelmann Reconsidered: The Evidence from Children Acquiring
Black English Vemaculen March 1978. (ERIC Document Reproduction Service No. ED 153 204, 31p.,
PC $3.32. MF $.83)
No. 83: Reyndds. R. E., Standiford, S. N.. & Anderson. R. C. Distribution of Reading Time When Questions
are Asked about a Restricted Category of Text Infomtation, April 1978. (ERIC Document Reproduc-
tion Service No. ED 153 206, 34p., PC $3.32. MF $.83)
No. 84: Baker. L Proofing Temporal Relationships in Simple Stories: Effects of Input Sequence, April
1978 (ERIC Document Reproduction Service No. ED 157 016, 54p.. PC K82, MF«$.83)
No. 85: Mason. J. M., Knisely. E.. & Kendall, J. Effects of Polysemous Words on Sentence Comprehen-
sion, May 1978 (EPIC Document Reproduction Service No. ED 157 015; 34p.. PC-$3.32. MF $.83)
No. 86: Anderson. T. H., Wardrop, J. L. Hively W., Muller, K. E.. Anderson. R. I.. Hastings. C. N.. &
Fredericksen. J. Development and Trie? of a Model for Developing Domain Referenced Tests of
Reading Comprehension, May 1978 (ERIC Document Reproduction Service No. ED 157 036, 69p..
PC $4.82. MF $.83)
No. 87: Andre. M. E. D. A.. & Anderson, T. H. The Development and Evaluation of a Self-Questioning
Study Technique, June 1978 (ERIC Document Reproduction Service No. ED 157 037. 37p.. PC $3.32.
MF-$.83)
No. 88: Bruce. B. C, & Newman. D. Interacting Plans, June 1978. (ERIC Document Reproduction Service
No. ED 157 038. AOOp., PC $6.32. Mr$.83)
No. 89: Bruce. B. C. Collins. A.. Rubin. A. D.. & Centner. D. A Cognitive Science Approach to Writing June
1978 (ERIC Document Reproduction Service No. ED 157 039. 57p.. PC $4.82. MF $.83)
No. 90: Asher. S. R. Referential Communication, June 1978. (ERIC Document Reproduction Service No.
ED 159 597, 71p.. PC-$4.82. MF $.83)
No. 91. Royer. J. M.. & Cunningham. D. J. On the Theory and Measurement of Reading Comprehension,
June 1978 (ERIC Document Repro^.uction Service No. ED 157 040. 63p.. PC-K82. MF-$.83)
No. 92: M3r:>n. J. M.. Kendall. J. R. Facilitating Reading Comprehension Through Text Structure Manipu-
lation, June 1978 (ERIC Document Reproduction Service No. ED 157 041. 36p.. PC-$3.32. MF-$.83)
No. 93: Ortony. A.. Schallert. D. L, Reynolds. R. E.. & Antos, S. J. Interpreting Mataphors and Idioms:
Some Effects of Context on Comprehension, July 1978. (ERIC Document Reproduction Service No.
ED 157 042. 41p.. PC.$3.32, MF-$.83)
No. 94: Brown. A. L, Campione. J. C. & Barclay, C. R. Training Self<:hecking Routines for Estimating
Test Readiness: Generalization from List Learning to Prose Recall, July 1978 (ERIC Document
Reproduction Service No. ED 158 226, 41 p.. PC-$3.32. ^^F $.83)
No. 95: Reichman. R. Conversational Coherency, July 1978. (ERIC Document Reproduction Service No.
ED 159 658. 86p., PC.$6.32. MF-$.83)
No. 96: Wigfield. A.. & Asher, S. R. Age Differences in Childrefn's Referential Communication Perfor-
mance: An Investigation of Task Effects, July 1978. (ERiC Document Reproduction Service No.
ED 159 659. 31p.. PC $3.32, MF-$.83)
No. 97: Steffensen. M. S.. Jogdeo. C. & Anderson. R. C. A Cross-Cultural Perspective on Reading
Comprehension, July 1978. (ERIC Document Reproduction Service No. ED 159 660, 41p.. PC-$3.32.
MF$.83)
No. 98: Green. G. M. Discourse Functions of Inversion Construction, July 1978. (ERIC Document Repro-
duction Service No. ED 160 998. 42p.. PC $3.32. MF-$.83)
No. 99: Asher. S. R. Influence of Topic Interest on Black Chiktren and White Chiklren's Reading
Comprehension, July 1978. (ERIC Document Reproduction Service No. ED 159 661, 35p.. PO$3.32.
MF.$.83)
No. 100: Jenkins. J. R.. Pany. D.. & Schreck. J. Vocabulary and Reading Comprehension: Instructional
Effects;, August 1978 (ERIC Document Reproduction Service No. ED 160 999. 50p.. PC $3.32. MF-
$.83)
No. 101: Shoben. E. J.. Rips. L J.. & Smith. E. E. Issues in Semantic Memory: A Response to Glass and
HolyoaK August 1978 (ERIC Document Reproduction Service No. ED 159 662. 85p.. PC $6.32. MF-
$.83)
No. 102: Baker. L. & Stein. N. L 77)e Development of Prose Comprehension Skills, September 1978
(ERIC Document Reproduction Service No. ED 159 663. 69p.. PC $4.82. MF*$.83)
No. 103: Fleisher. L $.. Jenkins. J. R.. & Pany. D. Effects on Poor Readers' Comprehension of Training in
Rapki Decoding. September 197a (ERIC Document Reproduction Service No. ED 159 664. 39p., PC-
$3.3a MF $.83)
i05
No. 104: Ariderson, T. H. Study Skills and Learning Strategies, September 1978. (ERIC Document Repro-
duction Service No. ED 161 000, 41p.. PC.$3.32. MF-$.83)
No. 105: Ortony. A. Beyond Uteral Similarity, October 1978. (ERIC Document Reproduction Service No
ED 166 635. 58p.. PC $4.82. MF $.83)
No. 106: Durkin. D. Wt^at Classroom Observations Reveal about Reading Comprehension Instruction,
October 1978. (ERIC Document Reproduction Service No ED 162 259. 94p.. PC $6.32. MF $.83)
No. 107: Adams. M. J. Models of Word Recognition, October 1978. (ERIC Document Rpproduction Ser-
vice No. ED 163 431. 93p.. PC-$6.32. MF.$.83)
No. 108: Reder. L M. Comprehension and Retention of Prose: A Literature Review, November 1978.
(ERIC Document Reproduction Service No. ED 165 114. 116p.. PC-$7.82. MF $.83)
No. 109: Wardrop. J. L. Anderson. T. H.. Hively. W.. Anderson. R. I.. Hastings. C. N.. & Muller. K. E. A Frame-
work for Analyzing Reading Test Characteristics, December 1978. (ERIC Document Reproduction
Service No. ED 165 117. 65p.. PC.$4.82. MF.$.83)
No. 110: Tirre. w. C. Manelis. L. & Leicht. K. L. The Effects of Imaginal and Verbal Strategies on Prose
Comprehension in Adults, December 1978. (ERIC Document Reproduction Servic? No ED 165 116
27p.. PC-$3.32. MF-$.83)
No. Ill Spiro. R. J.. & Tirre. W. C. Individual Differences in Schema Utilization During Discourse Pro-
cessing, January 1979. (ERIC Document Reproduction Service No. ED 166 651 29p PC-$3 32 MF-
$.83) .
No. 112: Ortony. A Some Psycholinguistic Asp^ts of Metaphor, January 1979. (ERIC Documer,t Repro-
duction Service No. ED 165 115. 38p.. PC $3.32. MF.$.83)
No. 313: Antos. S. J. Processing Fecilitation in a Lexical Decision Task, January 1979. (ERIC Document
Reproduction Service No. ED 165 129. 84p.. PC $6.32. MF-$.83)
No. 114: Centner D Semantic Integration at the Level of Verb Meaning, February 1979. (ERIC Docu-
ment Reproduction Service No. ED 165 130. 39p.. PC-$3.32. MF-$.83)
No. 115: Gearhart. M., & Hall. W. S. Internal State Words: Cultural and Situational Variation in Vocabu-
lary Usage, February 1979. (EPIC Document Repro^Juction Se.vice No ED 165 131 66p PC-$482
MF-$.83)
No. 116: Pearson. P. D.. Hansen. J.. & Gordon. C. The Effect of Background Knowledge on Young
Children's Comprehension of Explicit and Implicit Information, March 1979. (ERIC Document Rcpro
duction Service No. ED 169 521. 26p.. PC-$3.32. MF-$.83)
No. 117: Barnitz. J. G. Reading Cof ^prehension of Pronoun-Referent Structures by Children in Grades-
Two, Four, and Six, March 1979. (ERIC Document Reproduction Service No ED 170 731 51p PC-
$4.82. MF-$.83)
No. 118: Nicholson. T.. Pearson. P. D. & Dykstra. R. Effects of Embedded Anomalies and Oral Reading
Errors on Children's Understanding of Stories, March 1979. (ERIC Document Reproduction Service
No. ED 169 524. 43p.. PC $3.32. MF $.83)
No, 119: Anderson. R C. Pichert. J. W.. & Shirey. L L. Effects of the Reader's Schema at Different Points
in rime, April 1979. (ERIC Document Reproduction Service No. ED 169 523. 36p.. PC-$3.32. MF-$.83)
No. 120: Canney. G.. & Winograd. P. Schemata for Reading and Reading Comprehension Performance,
April 1979. (ERIC Document Reproduction Service No. ED 169 520. 99p.. PC-$6.32. MF-$.83)
No. Hall. W. S.. & Guthrie. L F. On the Dialect Question and Reading May 1979. (ERIC Document
Reproduction Service No. ED 169 522. 32p.. PC $3.32. MF $.83)
No. 122: McClure. E.. Mason. J.. & Barnitz. J. Story Structure and Age Effects o/t Children's Ability to
Sequence Stories, May 1979. (ERIC Document Reproduction Service No. ED 170 732. 75p PC-$482
MF-$.83)
No. 123: Kleiman. G. M.. Winograd. P. N.. & Humphrey. M. M. Prosody and Children's Parsing of Sen-
tences, May 1979. (ERIC Document Reproduction Service No. ED 170 733. 28p.. PC-$3.32, MF-$.83)
No. 124: Spiro. R. J. Etiology Of Reading Comprehension Style, May 1979. (ERIC Document Reproduction
Service No. ED 170 734. 21p.. PC-$1.82. MF-$.83)
No. 125: Hall. W. S.. & Tirre. W. C. The Communicative Environment of Young Children: Social Class,
Ethnic, and Situational Differences, May 1979. (ERIC Document Reproduction Service No'
ED 170 788. 30p.. PC-$3.32. MF.$.83)
No. 126: Mason. J.. & McCormick. C. Testing the Development of Reading and Linguistic Awareness,
May 1979. (ERIC Document Reproduction Service No. ED 170 735. 50p.. PC $3.32. MF $.83)
10-
No. 127: Brown, A, L, & Campione, J. C. Permissible Inferences from the Outcome of Training Studies in
Cognitive Developnwnt Research, May 1979. (ERIC Document Reproduction Service No ED 170 736,
34p.. PC.$3.32, MF.$.83)
No. 128: Brown, A. L, & French, L A. 7?»e Zone of Potential Development: Implications for Intelligence
Testing in the Year 2000, May 1979. (ERIC Document Reproduction Service No. ED 170 737, 46p.,
PC $3.32, MF.$.83)
No. 129: Nezworski, T., Stein, N. L, & Tratwsso, T. Siory Structure Versus Content Effects on Children's
Recall and Evaluative Inferences, June 1979. (ERIC Document Reproduction Service No ED 172 187,
49p., PC-$3.32, MF $.83)
No. 130: Bruce, B. Analysis of Interacting Plans as a Guide to the Understanding of Story Structure,
June 1979. (ERIC Document Reproduction Service No. ED 174 951, 43p., PC $3.32, MF $.83)
No. 131: Pearson, P. D., Raphael, T., TePaske, N., & Hyser, C. The Function of Metaphor in Children's
Recall of Expository Passages, July 1979. (ERIC Document Reproduction Service No. ED 174 950,
41p., PC $3.32, MF $.83)
No. 132: Green, G. M. Organization, Goals, and Comprehensibility in Narratives: Newswriting, a Case
Study, July 1979. (ERIC Document Reproduction Service No. ED 174 949, 66p., PC.$4.82, MF $.83)
No. 133: Kleiman, G. M. The Scope of Facilitation of Word Recognition from Single Word and Sentence
Frame Contexts, July 1979. (ERIC Document Reproduction Service No. ED 174 947 61p, PC-$4.82,
MF-$.83)
No. 134: McConkie, G. W., Hogaboam. T. W., Wolverton, G. S., Zola, D., & Lucas, p. A. Toward the Use of
Eye Movements in the Study of Language Processing August 1979. (ERIC Document Reproduction
Service No. ED 174 968, 48p., PC-$3.32, MF $.83)
No. 135: Schwartz, R. M. Levels of Processing: The Strategic Demands of Reading Comprehension,
August 1979. (ERIC Document Reproduction Service No. ED 177 471, 45p., PC-$3.32, MF $.83)
No. 136: Anderson, R. C, & Freetxxly, P. Vocabulary Knowledge, August 1979. (ERiC Document Repro-
duction Service No. ED 177 480, 71p., PC $4.82, MF $.83)
No. 137: Royer, J. M., Hastings, C. N.. & Hook, C. A Sentence Verification Technique for Measuring /?eaci-
ing Comprehension, August 1979. (ERIC Document Reproduction Service No. ED 176 234 34p., PC-
$3.32, MF $.83)
No. 138: Spiro, R. J. Prior Knowledge and Story Processing: Integration, Selection, and Variation,
August 1979. (ERIC Document Reproduction Service No. ED 176 235, 41 p., PC 3 32, MF $.83)
No. 139: Asher, S. R., & Wigfidd, A. Influence of Comparison Tr^ning on Children'^ Referential Commun^
ication, August 1979. (ERIC Document Reproduction Service No. ED 177 493, 42p., PC-$3.32, MF $.83)
No. 140: Alessi, S. M., Anderson, T. H., & Goetz, E. T. An Investigation of Lookbacks During Studying, Sep-
tember 1979. (ERIC Document Reproduction Service No. ED 177 494, 40p., PC $3.32, MF-$.83)
No. 141: Cohen, P. R., & Perrault, C. R. Elements of a Plan-Based Theory of Speech Acts, September
1979. (ERIC Document Reproduction Service No. ED 177 497, 76p., PC $6.32, MF $.83)
No. 142: Grueneich, R.. & Trabasso, T. The Story as Social Environment: Children's Comprehension and
Evaluation of Intentions and Consequences, September 1979. (ERIC Document Reproduction Service
No. ED 177 496, 56p., PC.$4.82, MF.$.63)
No. 143: Hsrmon, G. On the Discourse Structure of Direct Quotation, September 1979. (ERIC Document
♦ Reproduction Service No. ED 177 495, 46p., PC-$3.32, MF-$.83)
No. 144: Goetz, t T., Andsrson, R. C, & Schallert, D. L 77ie Representation of Sentences in Memory, Sep-
tember 1979. (ERIC Document Reproduction Service No. ED 177 527, 71p., PC $4.82, MF $.83)
No. 145: Baker, L Con^prehension Monitoring: Uentifying and Coping with Text Confusions, September
1979. (ERIC Document Reproduction Service No. ED 177 525, 62p., PC-$4.82, MF-$.83)
No. 146: Hall, W. S., & rtegy, W. E. Theoretical Issues in the lnvestigatk)n of Wbrds of Internal Report,
October 1979. (ERIC Document Reproduction Service No. ED 177 526, I08p., PC-$7.82, MF-$.83)
No. 147: Stein, N. L, & Goldman, S. Chiktren's Knowledge about Social Situations: From Causes to
Consequences, October 1.979. (ERIC Document Reproduction Service No. ED 177 524, 54p, PC $4.82,
MF-$.83)
No. 148: Hall, W. S., & Guthrie, L F. Cultural and Situational Variation in Language Function and Use:
Methods and Procedures for Research, October 1979. (ERIC Document Reproduction Service No
ED 179 944, 49p., PC-$3.3a MF-$.83)
No. 149; Pichert, J. W. Sensitivity to What is Important in Prose, November 1979. (ERIC Document
Reproduction Service No. ED 179 946, 64p., PC-K82, MF $.83)
IO7
No. 150: Dunn. B. R., Mathews. S. R.. il. & Bieger. G. Individual Differences in the Recall of Lower-Level
Textual Information. December 1979. (ERIC Document Reproduction Service No ED 181 M3, 37p
PC-$3.32, MF $.83)
No. 151: Gentner. D. Verb Semantic Structures in Memory for Sentences: Evidence for Componential
Representadon. December 1979. (ERIC Document Reproduction Service No. ED 181 424 75p PC-
$4.fe2. MF $.83)
No. 152: Tierney. R. J.. & Mosenthal. J. Discourse Comprehension and Production: Analyzing Text
Structure and Cohesion. January 1980. (ERIC Document Reproduction Service No ED 179 945. 84o
PC $6.32. MF $.83)
Nc. 153: Winograd, P.. & Johnston. P. Comprehmsicm Monitoring and the Error Detectioti Paradigm.
January 1980. (ERIC Document Reproduction Service No. ED 181 425. 57p.. PC K82. MF $.83>
No. 154: Ortony. A. Understanding Metaphors. January 1980. (ERiC Docur^ent Reproduction Service No
ED 181 426. 52p.. PC-$4.82. MF $.83)
No. 155: Anderson. T. H.. & Armbruster. B. B. Studying January 1980. (ERIC Document Reproduction
Service No. ED 181 427, 48p.. PC $3.32. MF $.83)
No. 156: Brown. A. L. & Ca.Tipione. J. C. Inducing Flexit)le Thinking: The Problem of Access. January
1980. (ERIC Document Reproduction Service No. ED 181 428. 44p.. PC $3.32. MF-$.83)
No. 157: Trabasso. T. On the Making of Inferences During Reading and Their Assessment. January
1980. (ERIC Documen* Reproduction Service No. ED 181 429. 38p.. PC-$3.32. MF-$.83)
No. 158: McClure. E.. & Steffensen. M. S. A Study of the Use of Conjunctions across Grades and Ethnic
Groups. January 1980. (ERIC Document Reproduction Service No. ED 182 688, 43p. PC-$332. MF-
$.83)
No. 159: Iran Nejad. A. 77je Schema: A Structural nr a Rmctional Pattarn. February 1280. (ER!C Docu
ment Reproduction Service Nc. ED 181 449. 46p.. PC $3.3?. MF $.83)
No. 160: Armbruster, B. E.. & Anderson. T. H. 7?»e Effect of Mapping on the Free Recall of Expository
Text. February 1980. (ERIC Document Reproduction Service No. ED 182 735. 49p.. PC $3.32. MF $.83;
No. 161: Hall. W. S.. & Dore. J. Lexical Sharing in Mother-Child Interaction. March 1980. (ERIC Document
Reproduction Service No. ED 184 066. 39p.. PC-$3.32. MF $.83)
No. 162: Davison. A.. Kantor. R. N.. Hannah. J.. Hermon. G.. Lutz. R.. Salziilo. R. Umita^s of Readability
Formulas in Guiding Adaptations of Texts. March 1980. (ERIC Document Reproduction Service No.
ED 184 090. 157p.. PC $10.82. MF-$.83)
No. 163: Unn. R. L. Levine. M. V.. Hastings. C. N.. & Wardrop. IL An Investigation of Item Bias in a Test
of Reading Comprehension. March 1980. (ERIC Document Reproduction Service No ED 184 091
97p.. PC-$6.32. MF-$.83)
No. 164: Seidenberg. M. S.. Tanenhaus. M. K.. & Le;man. J. M. The Time Course of Uixicel Ambiguity
Resolution in Context. March 1980. (ERIC Document Reproduction Service No. ED 184 092 58p PC-
$4.82. MF $.83) '
No. 165: Brown. A. L Learning and Development: The Problems of Compatibility. Access, end Induc-
tion. March 1980. (ERIC Document Reproduction Service No. ED 134 093. 76p.. PC-$6.32. MF $.83)
No. 166: Hansen. J.. & Pearson. P. D. The Effects of Inference Training and Practice on Young
Children's Comprehension. April 1980. (ERIC Document Reproduction Service No. ED 186 839 53p
PC-$4.82. MF $.83)
No. 167: Straker. D. Y. Situational Variables in Uinguaiie Use. April 1980. (ERIC Document Reproduction
Service No. ED 185 619. 49p.. PC $3.3a MF $.83)
No. 168: Green, G. M., Kantor, R. N.. Morgan. J. L, Stein. N. L. Hermon. G.. Salziilo. R.. Ssllner. M. B..
Bruce. B. C, Gentner. D.. & Wsbber, B. L Problems and Techniques of Text Analysis. April 19So!
(ERIC Document Reproduction Service Nc ED 185 513. 173p., PC-$10.82. MF $.83)
No. 169: Green. G. M., Kantor, R. N.. Morgan. J. L. Stein. N. L. Hermon, G.. Salziilo, R., & Sellner, M. B.
Analysis of Ba bar Loses His Crown . April 1980. (ERIC Document Reproduction Service No
ED 185 514. 89p.. PC-$6.3^ MF $.83)
No. 170: Green. G. M., Kantor. R N.. Morgan, J. L. Stein. N. L. Hermon. G.. Salziilo. R.. & SeJI.'-.er. M. B.
Analysis of "The Wonderful Desert." April 1980. (ERIC Document Reproduction Service No
ED 185 515, 47p., PC-$3.32, MF $.ai)
No. 171: Zehler, A. M., & Brewer. W. F. Acquisition of the Artich System in En^ish. May 1980. (ERIC
Document Reproduction Service No. ED 186 907. 51p.. PC $4.82. MF-$.83)
No. 172: Reynolds, R. E., & Ortony, A. Some Issues in the Measurement of Children's Comprehension of
Metaphorical Language, May 1980. (ERIC Document Reproduction Service No. ED 185 542. 42p.. PC-
$33Z MF $.83)
No. 173. Davison. A. Linguistics and the Measurement of Syntactic Complexity: The Case of Raising,
May 1980. (ERIC Document Reproduction Service No. ED 186 848. 60p.. PC $4.82. MF-$.83)
No. 174: Tirre. W. C. Freebody. P.. & Kaufman. K. Achit^ement Outcomes of Two Reading Programs:
An Instance of Aptitude-Treatment Interaction, June 1980.
No. 175: Asher. S. R.. & Wigfield. A Training Referential Con^munication Skills, July 1980.
No. 176: Tanenhaus. M. K.. & Seidenberg. M. S. Discourse Context and Sentence Perception, July 1980.
No. 177: Hall. W. S.. Unn. R. L. & Nagy. W. E. Spoken Words, August 1980.
No. 178: Tanenhaus. M. K.. Flanigan. H.. & Seidenberg. M. S. Orthographic and Phonological Activation in
Auditory and Visual Word Recognition, August 1980.
No. 179: Green, G. M. LJnguistics and the Pragmatics of Language Use: What You Know When You
Know a Language . . . and What Else You Know, August 1980.
No. 180: Steffensen. M. S,. & Guthrie. L F. Effect of Situation on the Vert)alization of Black Inner-City
Chikiren, September 1980.
No. 181: Green, G. M.. & Laff. M. 0. Five-Year Olds' Recognition of Authorship by Literary Style, Sep-
tember 1980.
No. 182: Collins, A. & Smith, E. E. Teaching the Process of Reading Comprehension, September 1980.
No. 183: Reynolds. R. E.. & Anderson. R. C. Influence of Questions on the Allocation of Attention during
Reading, October 1980.
No. 184: Iran-Nejad. A. Ortony. A.. & Rittenhouse. R. K. The Comprehension of Metaphorical Uses of
English by Deaf Children, October 1980.
No. 185: Smith, E. E. Organization of Factual Knowledge, October 1980.